Safety management system for building construction monitoring

Abstract

A safety management system for building construction monitoring disclosed by the present invention comprises a sensing layer, an edge layer and a platform layer, the sensing layer is used for full-dimensional data acquisition, and the sensing layer comprises a personnel safety monitoring module, an equipment state monitoring module, an environment monitoring module and a structure health monitoring module. The personnel safety monitoring module comprises an intelligent safety helmet and a positioning bracelet, the equipment state monitoring module comprises a tower crane monitoring subsystem and an elevator monitoring subsystem, the environment monitoring module comprises a gas detection array and a micro meteorological station, and the structure health monitoring module is composed of a fiber bragg grating sensing network. According to the safety management system for building construction monitoring, the problems of monitoring blind areas and data islands existing in traditional monitoring are effectively solved, a complete data base is provided for safety management, emergency time is shortened by more than 80% compared with traditional manual disposal, and the transmission reliability of key data is ensured while wide-area coverage is ensured.

Description

Technical Field

[0001] This invention relates to the field of building safety management, and in particular to a safety management system for monitoring building construction. Background Technology

[0002] In the field of construction safety management, with the expansion of project scale and the increase in process complexity, traditional safety management models have gradually exposed bottlenecks such as single monitoring methods, lagging risk prediction, and inefficient emergency response. Existing technological systems generally employ distributed monitoring equipment, such as independently operating positioning devices or fixed-point sensors. The lack of data exchange mechanisms between systems leads to fragmented safety information. Managers must rely on manual inspections and experience-based judgment to complete risk assessments, making it difficult to achieve real-time monitoring of dynamic hazards such as tower crane trajectories and foundation pit deformation, and even more difficult to quickly construct comprehensive emergency response plans in the event of emergencies.

[0003] The existing technical architecture has multiple limitations: the monitoring layer mostly uses single-parameter acquisition devices, such as video surveillance or fixed sensors to obtain local data, which cannot cover three-dimensional work spaces and is difficult to capture the correlation between personnel physiological state and equipment operating parameters; the data analysis link relies heavily on manual statistics and post-event review, which not only makes it difficult to extract risk signs from massive amounts of data, but is also prone to misjudgment or omission due to subjective judgment; at the emergency response level, traditional early warning mechanisms can only trigger simple audible and visual alarms, which cannot automatically match the response strategy according to the risk level, nor can they realize the immediate issuance of automated control commands such as dust spraying and equipment braking, resulting in the timeliness and execution accuracy of emergency response remaining at a low level for a long time. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides a safety management system for building construction monitoring, which solves the aforementioned technical problem of difficulty in instantly issuing automated control commands such as dust suppression spraying and equipment braking, resulting in consistently low emergency response timeliness and execution accuracy.

[0005] To achieve the above objectives, this application provides the following technical solution: a safety management system for building construction monitoring, comprising a perception layer, a decision layer, an edge layer, and a platform layer. The perception layer is used for full-dimensional data acquisition and includes a personnel safety monitoring module, an equipment status monitoring module, an environmental monitoring module, and a structural health monitoring module. The personnel safety monitoring module includes a smart safety helmet and a positioning wristband. The equipment status monitoring module includes a tower crane monitoring subsystem and an elevator monitoring subsystem. The environmental monitoring module includes a gas detection array and a micro-weather station. The structural health monitoring module is composed of a fiber optic grating sensor network.

[0006] The edge layer is used for real-time intelligent analysis. The edge layer includes edge computing nodes, a cloud security analysis platform, and an AI risk assessment engine. The cloud security analysis platform includes a digital twin engine and a big data platform.

[0007] The decision-making layer is used for closed-loop safety management and control. The decision-making layer includes a dynamic early warning mechanism, an automated response module, and digital twin visualization. The dynamic early warning mechanism consists of level one to level four early warnings. The automated response module includes equipment linkage control and robot collaboration. The equipment linkage control uses the Modbus TCP protocol to communicate with the PLC controller to realize the automatic start of the sprinkler system in fire scenarios and the locking of lifting machinery around the foundation pit in the event of a collapse.

[0008] The perception layer and edge layer upload sensor data to the edge node via LoRaWAN. The edge layer transmits information to the cloud via TCP and IP protocols. The cloud transmits data to the BIM collaboration platform and then to the application layer via RESTful API.

[0009] Preferably, the smart safety helmet integrates a UWB chip, a six-axis gyroscope, and physiological electrode pads for sub-meter level positioning, attitude recognition, and physiological monitoring. The positioning wristband is equipped with an emergency call button to trigger an SOS signal. The UWB chip built into the smart safety helmet uses the TDOA positioning algorithm to achieve sub-meter level accuracy. The six-axis gyroscope uses an attitude calculation algorithm to identify abnormal behaviors such as falls and impacts. The physiological electrode pads collect brainwave and heart rate variability data in real time, and work with an edge AI model to achieve fatigue state early warning.

[0010] Preferably, the tower crane monitoring subsystem is equipped with a slewing mechanism and hoisting wire rope in the tower crane, and is equipped with vibration sensors and tilt sensors. The hoist monitoring subsystem consists of a load sensor and is linked with the door magnetic switch in the tower crane to prohibit the operation of the tower crane under overload and door-open conditions. The vibration sensor of the tower crane monitoring subsystem uses a triaxial accelerometer to monitor the dynamic characteristics of the slewing mechanism, and the tilt sensor uses Kalman filtering to compensate for installation errors. The load sensor of the hoist monitoring subsystem uses a strain gauge measurement principle. The door magnetic switch and the tower crane control system are hardware-level interlocked to ensure that the main contactor power is directly cut off in the overload or door-open state.

[0011] Preferably, the gas detection array deploys electrochemical sensors to monitor CO, H2S, and O2 concentrations. The micro-weather station integrates temperature and humidity, wind speed and direction, and rainfall sensors, and the data is used to predict collapse and fire risks. The electrochemical sensors of the gas detection array adopt the constant potential electrolysis principle. The CO sensor has a range of 0-1000ppm, the H2S sensor has a design to resist sulfide poisoning, and the O2 sensor adopts a current-limiting diffusion electrode. The wind speed sensor of the micro-weather station is measured by ultrasonic time difference method, and the rainfall data is acquired through a piezoelectric rain gauge. All parameters are input into an LSTM-based collapse prediction model for dynamic risk assessment.

[0012] Preferably, the structural health monitoring module, in conjunction with optical fibers deployed along the foundation pit support piles and main structural beams and columns, achieves strain monitoring and temperature compensation. Simultaneously, the data is transmitted to edge nodes via wavelength division multiplexing (WDM) technology. The fiber optic grating sensor network employs distributed Brillouin scattering technology to achieve long-distance strain monitoring. The foundation pit support piles are equipped with a dense sensor array with a spacing of 50cm. The main structural beams and columns utilize wavelength encoding to achieve multi-parameter multiplexing. The WDM module achieves 128-channel parallel demodulation through an arrayed waveguide grating. Data transmission employs the SSH encryption protocol to ensure communication security.

[0013] Preferably, the edge computing node is equipped with an NVIDIA Jetson AGX Xavier module and supports 5G / Wi-Fi 6 dual-mode communication. The processing flow includes data cleaning to remove sensor noise, feature extraction, extraction of RMS and peak factor features from the time and frequency domains, local decision-making, and running of lightweight MobileNetV3 to achieve preliminary risk assessment. The NVIDIA Jetson AGX Xavier module on the edge computing node is equipped with a 512-core Volta GPU, and the 5G module supports SA / NSA dual-mode networks. The data cleaning process includes wavelet threshold denoising and moving average filtering. The feature extraction module calculates the time-domain RMS value, peak factor, and frequency-domain power spectral density. The MobileNetV3 model is compressed to 8MB in size through knowledge distillation technology, and the inference speed reaches 30fps.

[0014] Preferably, the digital twin engine is based on Unity 3D to build a BIM model of the construction site, supports data mapping and historical playback, the big data platform is used to store structured and unstructured data, adopts a Hadoop+HBase distributed architecture, supports PB-level data storage, the digital twin engine is based on the Unity 3D entity component system architecture to achieve efficient rendering of hundreds of thousands of components, the BIM model supports IFC4.1 standard data exchange, the historical playback function uses a spatiotemporal index tree to accelerate data retrieval, the Hadoop cluster of the big data platform is configured with 32 Datanode nodes, and the HBase table design uses a spatiotemporal partition key to optimize range query performance.

[0015] Preferably, the model architecture of the AI ​​risk assessment engine adopts an LSTM-Transformer hybrid model. The input is multi-source time-series data, and the training data covers no less than 3 years of historical accident data. The data annotation adopts a semi-supervised learning method. The LSTM layer of the AI ​​risk assessment engine captures temporal dependencies, the Transformer encoder extracts spatial correlation features, and the hybrid model dynamically fuses dual-stream features through a gating mechanism. The training data contains 2 million labeled samples, of which 10% are labeled by experts through active learning. The model optimization adopts knowledge distillation technology to improve deployment efficiency.

[0016] Preferably, the Level 1 warning is triggered when the tower crane vibration amplitude is greater than 0.5g and lasts for 10 minutes, pushing the message to the safety officer's APP and requiring on-site verification; the Level 2 warning is triggered when the foundation pit settlement rate is greater than 2mm / d, activating an audible and visual alarm and restricting surrounding operations; the Level 3 warning is triggered when the gas concentration exceeds the standard, cutting off the power to hazardous equipment and broadcasting evacuation instructions; and the Level 4 warning is triggered when the structural strain is greater than 80% of the ultimate bearing capacity, linking the emergency system and directing all personnel to evacuate. Warning propagation is achieved through the WebSocket protocol, supporting SMS, APP, and broadcast push. After the Level 1 warning is triggered, a verification work order is automatically created through the enterprise WeChat robot; the Level 2 warning is linked to the on-site IP broadcast system to play customized voice; the Level 3 warning cuts off the power using a solid-state relay to achieve millisecond-level response; and the Level 4 warning evacuation instructions are broadcast regionally through a UWB base station. The WebSocket protocol supports a heartbeat mechanism to ensure reliable message transmission.

[0017] Preferably, the robot collaboration includes deploying a quadrupedal inspection robot equipped with an infrared thermal imager and a gas detector to replace manual entry into enclosed spaces for investigation. The digital twin visualization supports immersive inspection via VR headsets, retrieves sensor data through gesture recognition, and provides a risk heatmap overlay function. High-risk areas are marked with flashing red. The quadrupedal inspection robot is equipped with a dual-light gimbal to achieve 360° environmental perception. Path planning uses an improved algorithm to avoid dynamic obstacles. The VR visualization system achieves cross-platform compatibility through the SteamVR plugin. Gesture recognition uses the MediaPipe framework to track 21 hand joints in real time. The risk heatmap is dynamically generated based on a kernel density estimation algorithm.

[0018] In summary, this application provides a safety management system for building construction monitoring, which has the following beneficial effects: This construction safety management system achieves intelligent closed-loop management of construction safety through a three-layer architecture. The perception layer uses a combination of multimodal monitoring modules, and smart safety helmets and positioning wristbands can simultaneously acquire personnel physiological parameters and spatial location information. The tower crane monitoring subsystem captures the equipment operating status in real time through load sensors and inclinometers, while the fiber optic grating sensor network achieves high-precision capture of structural stress changes with distributed measurement characteristics. This effectively solves the problems of monitoring blind spots and data silos in traditional monitoring, providing a complete data foundation for safety management.

[0019] The real-time analysis system built on the edge layer preprocesses raw data through edge computing nodes and combines it with the 3D modeling capabilities of the cloud-based digital twin engine to achieve dynamic mapping and risk simulation of construction scenarios. The AI ​​risk assessment engine, based on a prediction model trained on historical data, can identify hidden risks such as tower crane collision trajectories and foundation pit deformation trends. Compared with manual inspection, it improves risk identification efficiency while keeping the false alarm rate within a reasonable threshold. The four-level early warning mechanism automatically triggers differentiated response strategies according to the risk level. When the concentration of harmful gases exceeds the standard, the system can immediately send control commands to the PLC via the Modbus TCP protocol to start the sprinkler system within 300 milliseconds, shortening the emergency response time by more than 80% compared to traditional manual handling.

[0020] The decision-making layer's automated response module innovatively integrates equipment linkage and robot collaboration technologies. In scenarios involving pre-collapse monitoring, the system can not only lock lifting machinery around the foundation pit but also dispatch inspection robots for localized reinforcement work. The digital twin visualization platform, through the fusion of BIM models and real-time data, enables managers to intuitively grasp the risk propagation path and assist in formulating precise control plans. The entire system adopts a hybrid LoRaWAN and TCP / IP networking approach, ensuring wide-area coverage while guaranteeing the reliability of critical data transmission. Attached Figure Description

[0021] Figure 1 This is a system structure diagram of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] This application provides a technical solution; please refer to [link / reference]. Figure 1 The safety management system for building construction monitoring includes a perception layer, a decision layer, an edge layer, and a platform layer. The perception layer achieves full-dimensional data acquisition through a multi-modal sensor network. The personnel safety monitoring module adopts a combination monitoring scheme of smart safety helmets and positioning wristbands. The equipment status monitoring module designs differentiated monitoring subsystems for tower cranes and elevators. The environmental monitoring module integrates real-time sensing of gas and meteorological parameters. The structural health monitoring module uses a fiber optic grating sensor network to achieve submicron-level deformation monitoring.

[0024] The perception layer is used for full-dimensional data collection. The perception layer includes a personnel safety monitoring module, an equipment status monitoring module, an environmental monitoring module, and a structural health monitoring module. The personnel safety monitoring module includes a smart safety helmet and a positioning wristband. The equipment status monitoring module includes a tower crane monitoring subsystem and an elevator monitoring subsystem. The environmental monitoring module includes a gas detection array and a micro-weather station. The structural health monitoring module is composed of a fiber optic grating sensor network. The edge layer constructs a low-latency analysis network through distributed edge computing nodes, the cloud security analysis platform realizes virtual-real mapping based on a digital twin engine, and the AI ​​risk assessment engine adopts a spatiotemporal attention mechanism to enhance the multi-source data fusion capability.

[0025] The edge layer is used for real-time intelligent analysis. The edge layer includes edge computing nodes, a cloud security analysis platform, and an AI risk assessment engine. The cloud security analysis platform includes a digital twin engine and a big data platform. The decision-making layer constructs a three-level closed-loop control mechanism: a dynamic early warning mechanism triggers differentiated responses through graded thresholds, an automated response module realizes cascaded control of equipment and autonomous decision-making by robots, and digital twin visualization provides three-dimensional spatial risk heat maps and emergency plan simulation functions.

[0026] The decision-making layer is used for closed-loop safety management and control. The decision-making layer includes a dynamic early warning mechanism, an automated response module, and digital twin visualization. The dynamic early warning mechanism consists of level one to level four early warnings. The automated response module includes equipment linkage control and robot collaboration. The equipment linkage control uses the Modbus TCP protocol to communicate with the PLC controller to realize the automatic start of the sprinkler system in fire scenarios and the locking of lifting machinery around the foundation pit in the event of collapse. The data transmission link adopts a hybrid networking scheme: the perception layer uses LoRaWAN to achieve wide-area low-power coverage, the edge layer uses 5G slicing technology to ensure the priority of key data transmission, and the cloud uses RESTful API to realize the spatiotemporal alignment of BIM models and monitoring data, thus building a data channel for full-element interconnection.

[0027] The perception layer and edge layer upload sensor data to the edge node via LoRaWAN. The edge layer transmits information to the cloud via TCP and IP protocols. The cloud transmits data to the BIM collaboration platform and then to the application layer via RESTful API.

[0028] The smart safety helmet integrates a UWB chip, a six-axis gyroscope, and physiological electrode pads for sub-meter-level positioning, attitude recognition, and physiological monitoring. The positioning wristband is equipped with an emergency call button to trigger an SOS signal. The UWB chip built into the smart safety helmet uses the TDOA positioning algorithm to achieve sub-meter-level accuracy. The six-axis gyroscope uses an attitude calculation algorithm to identify abnormal behaviors such as falls and impacts. The physiological electrode pads collect brainwave and heart rate variability data in real time, and work with an edge AI model to provide fatigue warnings.

[0029] The tower crane monitoring subsystem is equipped with a slewing mechanism and hoisting wire rope in the tower crane, and is equipped with vibration sensors and tilt sensors. The hoist monitoring subsystem consists of a load sensor and is linked with the door magnetic switch in the tower crane to prevent the tower crane from operating under overload or door-open conditions. The vibration sensor of the tower crane monitoring subsystem uses a triaxial accelerometer to monitor the dynamic characteristics of the slewing mechanism, and the tilt sensor uses Kalman filtering to compensate for installation errors. The load sensor of the hoist monitoring subsystem uses a strain gauge measurement principle. The door magnetic switch and the tower crane control system are hardware-level interlocked to ensure that the main contactor power is directly cut off in the overload or door-open condition.

[0030] The gas detection array deploys electrochemical sensors to monitor CO, H2S, and O2 concentrations. The micro-weather station integrates temperature, humidity, wind speed and direction, and rainfall sensors, and the data is used to predict collapse and fire risks. The electrochemical sensors of the gas detection array adopt the constant potential electrolysis principle. The CO sensor has a range of 0-1000ppm, the H2S sensor has a design to resist sulfide poisoning, and the O2 sensor uses a current-limiting diffusion electrode. The wind speed sensor of the micro-weather station uses the ultrasonic time-of-flight method for measurement, and the rainfall data is acquired through a piezoelectric rain gauge. All parameters are input into an LSTM-based collapse prediction model for dynamic risk assessment.

[0031] The structural health monitoring module, in conjunction with optical fibers deployed along the foundation pit support piles and main structural beams and columns, enables strain monitoring and temperature compensation. Simultaneously, data is transmitted to edge nodes via wavelength division multiplexing (WDM). The fiber optic grating sensor network employs distributed Brillouin scattering technology for long-distance strain monitoring. The foundation pit support piles are equipped with a dense sensor array spaced at 50cm intervals. Wavelength encoding is used in the main structural beams and columns to achieve multi-parameter multiplexing. The WDM module achieves 128-channel parallel demodulation through an arrayed waveguide grating. Data transmission uses the SSH encryption protocol to ensure communication security.

[0032] The edge computing node is equipped with an NVIDIA Jetson AGX Xavier module and supports 5G / Wi-Fi 6 dual-mode communication. The processing flow includes data cleaning to remove sensor noise, feature extraction, extraction of RMS and peak factor features from the time and frequency domains, local decision-making, and running a lightweight MobileNetV3 to achieve preliminary risk assessment. The NVIDIA Jetson AGX Xavier module on the edge computing node is equipped with a 512-core Volta GPU, and the 5G module supports SA / NSA dual-mode networks. The data cleaning process includes wavelet threshold denoising and moving average filtering. The feature extraction module calculates the time-domain RMS value, peak factor, and frequency-domain power spectral density. The MobileNetV3 model is compressed to 8MB in size through knowledge distillation technology, and the inference speed reaches 30fps.

[0033] The digital twin engine is based on Unity 3D to build a BIM model of the construction site, supporting data mapping and historical playback. The big data platform is used to store structured and unstructured data, adopting a Hadoop+HBase distributed architecture to support petabyte-level data storage. The digital twin engine, based on the Unity 3D entity component system architecture, achieves efficient rendering of hundreds of thousands of components. The BIM model supports IFC4.1 standard data exchange. The historical playback function uses a spatiotemporal index tree to accelerate data retrieval. The Hadoop cluster of the big data platform is configured with 32 Datanode nodes, and the HBase table design uses a spatiotemporal partition key to optimize range query performance.

[0034] The AI ​​risk assessment engine adopts an LSTM-Transformer hybrid model architecture. The input is multi-source time-series data, and the training data covers no less than 3 years of historical accident data. Data annotation adopts a semi-supervised learning method. The LSTM layer of the AI ​​risk assessment engine captures temporal dependencies, and the Transformer encoder extracts spatial correlation features. The hybrid model dynamically fuses dual-stream features through a gating mechanism. The training data contains 2 million labeled samples, of which 10% are labeled by experts through active learning. The model optimization adopts knowledge distillation technology to improve deployment efficiency.

[0035] The Level 1 warning is triggered when the tower crane vibration amplitude exceeds 0.5g for 10 minutes, pushing the message to the safety officer's APP and requiring on-site verification. The Level 2 warning is triggered when the foundation pit settlement rate exceeds 2mm / d, activating an audible and visual alarm and restricting surrounding operations. The Level 3 warning is triggered when the gas concentration exceeds the standard, cutting off the power to hazardous equipment and broadcasting evacuation instructions. The Level 4 warning is triggered when the structural strain exceeds 80% of the ultimate bearing capacity, linking the emergency system and directing all personnel to evacuate. Warning propagation is achieved through the WebSocket protocol, supporting SMS, APP, and broadcast push. After a Level 1 warning is triggered, a verification work order is automatically created through the enterprise WeChat robot. The Level 2 warning is linked to the on-site IP broadcast system to play customized voice messages. The Level 3 warning uses solid-state relays to achieve millisecond-level response for power cutting off. The Level 4 warning evacuation instructions are broadcast regionally through UWB base stations. The WebSocket protocol supports a heartbeat mechanism to ensure reliable message transmission.

[0036] The robot collaboration includes deploying a quadrupedal inspection robot equipped with an infrared thermal imager and a gas detector to replace manual labor in entering enclosed spaces for investigation. The digital twin visualization supports immersive inspection via VR headsets, retrieves sensor data through gesture recognition, and provides a risk heatmap overlay function. High-risk areas are marked with flashing red. The quadrupedal inspection robot is equipped with a dual-light gimbal to achieve 360° environmental perception. Path planning uses an improved algorithm to avoid dynamic obstacles. The VR visualization system achieves cross-platform compatibility through the SteamVR plugin. Gesture recognition uses the MediaPipe framework to track 21 hand joints in real time. The risk heatmap is dynamically generated based on a kernel density estimation algorithm.

[0037] The perception layer acquires raw data through a multimodal sensor network. In the personnel safety monitoring module, the UWB chip of the smart safety helmet uses the TDOA positioning algorithm to achieve sub-meter accuracy positioning. The six-axis gyroscope monitors the wearer's head movement trajectory in real time through a quaternion attitude calculation algorithm. When an abnormal fall posture is detected, the local decision-making module of the edge computing node is triggered. The emergency call button of the positioning wristband directly wakes up the LoRaWAN module through GPIO interrupt, encapsulating the SOS signal and the current coordinates into an emergency message. The equipment status monitoring module deploys a three-dimensional accelerometer to monitor the vibration characteristics of the slewing mechanism of the tower crane. A Kalman filter algorithm is used to compensate for installation errors in the tilt sensor data. The elevator load sensor converts mechanical deformation into voltage signals through a strain gauge bridge circuit. The door magnetic switch status is transmitted to the PLC controller after electrical isolation through an optocoupler. The electrochemical sensor array of the environmental monitoring module adopts the constant potential electrolysis principle. The CO sensor maintains a 0.4V operating voltage through a three-electrode system, the H2S sensor is equipped with a noble metal catalyst layer to prevent poisoning, and the O2 sensor adopts a current-limiting diffusion electrode structure. The micro-weather station's wind speed sensor measures airflow velocity using the ultrasonic time-of-flight method, while rainfall data is obtained by converting the charge generated by raindrop impacts on a piezoelectric ceramic plate. The structural health monitoring module deploys fiber optic gratings at 50cm intervals along the foundation pit support piles. Wavelength encoding technology is used to decouple temperature and strain parameters in the main structural beams and columns. The wavelength division multiplexing module achieves 128-channel parallel demodulation through an array of waveguide gratings.

[0038] The data acquisition terminal selects differentiated transmission channels based on data type. Personnel positioning data and equipment status parameters are uploaded via LoRaWAN network using spread spectrum modulation, and an adaptive rate algorithm dynamically adjusts SF7-SF12 parameters based on channel quality. Environmental monitoring data and structural health data are transmitted via dedicated bearers established using 5G slicing technology, utilizing network slicing selection functionality to achieve GBR transmission. After all data is aggregated at edge nodes, the edge computing platform timestamps the time-series data, encapsulates data packets using the SSH encryption protocol, and transmits them to the cloud-based big data platform via the TCP / IP protocol stack. Spatiotemporal alignment between the BIM model and monitoring data is achieved through a RESTful API. The coordinate system transformation module maps the sensor's latitude and longitude coordinates to the local coordinate system of the BIM model, and the time axis synchronization engine uses the NTP protocol to ensure the temporal consistency of multi-source data.

[0039] Edge computing nodes deployed on the construction site execute real-time analysis processes. The data cleaning module first applies a wavelet threshold denoising algorithm to filter out high-frequency noise, and a moving average filter smooths the vibration signal. The feature extraction engine calculates the RMS value and peak factor in the time domain and obtains the power spectral density in the frequency domain through a fast Fourier transform. A lightweight MobileNetV3 model performs preliminary risk assessment on the processed features. This model uses knowledge distillation technology to use the feature map of the teacher network as a supervisory signal, compressing the parameter size to 8MB while maintaining 92% accuracy. The edge nodes synchronously run an equipment status monitoring rule engine. When the tower crane vibration amplitude exceeds the 0.5g threshold, an early warning message is immediately sent to the safety officer's terminal via the MQTT protocol; when the hoist load exceeds 90% of the rated value, a Modbus TCP command is directly sent to the PLC controller to cut off the power.

[0040] A cloud-based security analysis platform constructs a digital twin to achieve virtual-real mapping. The Unity 3D engine loads an IFC4.1 standard BIM model, achieving efficient rendering of hundreds of thousands of components through a solid component system architecture. The historical data playback module adopts a spatiotemporal index tree structure, supporting rapid retrieval of monitoring records by time range and spatial region. The big data platform stores structured data based on a Hadoop+HBase distributed architecture, employing a column-family storage design to optimize time-series data query performance. Unstructured data, such as video streams, is stored in a distributed manner via HDFS. The AI ​​risk assessment engine's LSTM-Transformer hybrid model receives feature sequences uploaded from the edge layer. The LSTM layer captures temporal dependencies, the Transformer encoder extracts spatial correlation features through a multi-head attention mechanism, and the gating fusion unit dynamically adjusts the dual-stream weights based on the input data. The model training data contains 2 million labeled samples, 10% of which are labeled by security experts through an active learning strategy, and the remainder uses semi-supervised learning to generate pseudo-labels.

[0041] The dynamic early warning mechanism implements tiered responses based on risk levels. When a Level 1 warning is triggered, the WeChat robot automatically creates a verification work order containing the equipment number, abnormal parameters, and on-site photos, and pushes it to the safety officer's app via WebSocket. A Level 2 warning activates the on-site IP broadcast system, playing customized voice commands using the G.722 encoding standard, and simultaneously sending a speed limit command to the tower crane control system via Modbus TCP. Under a Level 3 warning, solid-state relays cut off power to hazardous equipment within 50ms, and gas detection data is broadcast to all connected terminals via WebSocket. When a Level 4 warning is triggered, the UWB base station sends an evacuation command to the designated area, and the digital twin engine generates the optimal escape route and overlays it onto the VR visualization interface. After receiving the task command, the quadrupedal inspection robot in the automated response module plans the inspection path based on an improved A* algorithm, and the dual-light gimbal transmits infrared thermal imaging and gas concentration data to the command center in real time. The digital twin visualization system supports gesture recognition interaction. Safety officers can retrieve detailed sensor data by tracking 21 hand joints through the MediaPipe framework. The risk heat map is dynamically rendered using a kernel density estimation algorithm, and high-risk areas are marked with a red gradient flashing.

[0042] After the emergency response is completed, the system automatically initiates the event debriefing process. The big data platform extracts monitoring data from HBase for one hour before and after the event, and uses spatiotemporal correlation analysis to locate the risk evolution chain. The digital twin engine loads the event snapshot for process playback, and the 3D visualization interface simultaneously displays the equipment status change curves and personnel movement trajectories. The AI ​​risk assessment engine incrementally learns from the event data and updates the weight parameters of the LSTM-Transformer model. The debriefing report generation module automatically summarizes key indicators such as early warning response timeliness, number of equipment linkages, and personnel evacuation routes, and uses NLP technology to convert structured data into natural language descriptions, ultimately generating a PDF document containing improvement suggestions that is pushed to management's email address.

[0043] This system achieves a complete closed loop of data collection, transmission, analysis, and decision-making through an edge-cloud collaborative architecture, with modules at each level working closely together to form a security protection network. From sub-meter-level personnel positioning to sub-micron-level structural deformation monitoring, from millisecond-level device linkage to minute-level risk early warning, it constructs a safety management system covering all elements of building construction, significantly improving risk prevention and control capabilities and emergency response efficiency.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0045] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A safety management system for construction monitoring, comprising a perception layer, a decision-making layer, an edge layer, and a platform layer, characterized in that: The perception layer is used for full-dimensional data collection. The perception layer includes a personnel safety monitoring module, an equipment status monitoring module, an environmental monitoring module, and a structural health monitoring module. The personnel safety monitoring module includes a smart safety helmet and a positioning wristband. The equipment status monitoring module includes a tower crane monitoring subsystem and an elevator monitoring subsystem. The environmental monitoring module includes a gas detection array and a micro-weather station. The structural health monitoring module is composed of a fiber optic grating sensor network. The edge layer is used for real-time intelligent analysis. The edge layer includes edge computing nodes, a cloud security analysis platform, and an AI risk assessment engine. The cloud security analysis platform includes a digital twin engine and a big data platform. The decision-making layer is used for closed-loop safety management and control. The decision-making layer includes a dynamic early warning mechanism, an automated response module, and digital twin visualization. The dynamic early warning mechanism consists of level one to level four early warnings. The automated response module includes equipment linkage control and robot collaboration. The equipment linkage control uses the Modbus TCP protocol to communicate with the PLC controller. The perception layer and edge layer upload sensor data to the edge node via LoRaWAN. The edge layer transmits information to the cloud via TCP and IP protocols. The cloud transmits data to the BIM collaboration platform and then to the application layer via RESTful API.

2. The safety management system for construction monitoring according to claim 1, characterized in that: The smart safety helmet integrates a UWB chip, a six-axis gyroscope, and physiological electrode pads for sub-meter level positioning, attitude recognition, and physiological monitoring. The positioning wristband is equipped with an emergency call button to trigger an SOS signal.

3. The safety management system for construction monitoring according to claim 1, characterized in that: The tower crane monitoring subsystem is equipped with a slewing mechanism and hoisting wire rope in the tower crane, and is equipped with vibration sensors and tilt sensors. The hoist monitoring subsystem consists of load sensors and is linked with the door magnetic switch in the tower crane to prohibit the operation of the tower crane under overload and door-open conditions.

4. The safety management system for construction monitoring according to claim 1, characterized in that: The gas detection array deploys electrochemical sensors to monitor CO, H2S, and O2 concentrations. The micro-weather station integrates sensors for temperature, humidity, wind speed and direction, and rainfall, and the data is used to predict collapse and fire risks.

5. The safety management system for construction monitoring according to claim 1, characterized in that: The structural health monitoring module, in conjunction with optical fibers laid along the foundation pit support piles and main structural beams and columns, enables strain monitoring and temperature compensation. The data is transmitted to edge nodes via wavelength division multiplexing technology.

6. The safety management system for construction monitoring according to claim 1, characterized in that: The edge computing node is equipped with an NVIDIA Jetson AGX Xavier module and supports 5G / Wi-Fi 6 dual-mode communication. The processing flow is data cleaning to remove sensor noise and extract features. It extracts RMS and peak factor features from the time and frequency domains and makes local decisions. It runs lightweight MobileNetV3 to achieve preliminary risk assessment.

7. The safety management system for construction monitoring according to claim 1, characterized in that: The digital twin engine is based on Unity 3D to build a BIM model of the construction site, supports data mapping and historical playback, and the big data platform is used to store structured and unstructured data. It adopts a Hadoop+HBase distributed architecture and supports petabyte-level data storage.

8. The safety management system for construction monitoring according to claim 1, characterized in that: The AI ​​risk assessment engine adopts an LSTM-Transformer hybrid model architecture, with multi-source time series data as input, training data covering no less than 3 years of historical accident data, and data annotation using a semi-supervised learning method.

9. The safety management system for construction monitoring according to claim 1, characterized in that: The first-level warning is triggered when the tower crane vibration amplitude is greater than 0.5g for 10 minutes, pushing the message to the safety officer's APP and requiring on-site verification. The second-level warning is triggered when the foundation pit settlement rate is greater than 2mm / d, activating the audible and visual alarm, restricting surrounding operations, and triggering the third-level warning. When the gas concentration exceeds the standard, the power supply to the hazardous equipment is cut off, evacuation instructions are broadcast, and the fourth-level warning is triggered. When the structural strain is greater than 80% of the ultimate bearing capacity, the emergency system is activated, and all personnel are directed to evacuate. Warnings are disseminated via the WebSocket protocol and support SMS, APP, and broadcast push notifications.

10. The safety management system for construction monitoring according to claim 1, characterized in that: The robot collaboration includes deploying a quadruped inspection robot equipped with an infrared thermal imager and a gas detector. The digital twin visualization supports immersive inspection with a VR headset, retrieves sensor data through gesture recognition, and provides a risk heat map overlay function. High-risk areas are marked with flashing red.

Images