Construction safety penetrating management multi-level linkage early warning system

Abstract

The present application relates to the technical field of construction safety management, and discloses a construction safety penetration type management multi-stage linkage early warning system.The system comprises a management layer, an execution layer and a work layer node.The management layer generates a global safety strategy, the execution layer decomposes and issues tasks, and the work layer collects on-site multi-modal safety data, including equipment vibration characteristics, environmental image flow, personnel positioning trajectory and safety operation records.The system first analyzes the changes in equipment vibration characteristics, divides stable and abnormal working condition periods, and selects high-risk areas according to the difference amount;in each high-risk area, the safety state deviation of the work layer node is calculated in combination with the task identifier and the environmental image flow;then, the responsibility trace weight is generated according to the spatio-temporal correlation and deviation of the personnel trajectory and operation record in the abnormal period;finally, the root node is selected according to the weight, multi-modal data is associated to construct a safety risk propagation tree, and multi-stage linkage precise safety management is realized.

Description

Technical Field

[0001] This invention relates to the field of construction safety management technology, specifically a multi-level linkage early warning system for construction safety penetration management. Background Technology

[0002] In fields such as construction, mining, and large-scale engineering projects, safety management is always a crucial aspect of ensuring the smooth progress of construction. Traditional construction safety management models rely heavily on manual inspections, periodic checks, and hierarchical information transmission. When faced with complex and ever-changing construction sites, this model often suffers from problems such as information lag, inefficient transmission, and blind spots in supervision.

[0003] As construction scale continues to expand, the types of equipment involved are increasing, the number of workers is rising, and the construction site environment is becoming more complex. Subtle anomalies in equipment operation, if not detected in time, can lead to serious safety accidents; workers' violations or entry into hazardous areas also pose significant risks to construction safety. However, most existing management systems can only achieve single-dimensional monitoring, such as collecting data on equipment vibration or tracking personnel location, making it difficult to effectively integrate and analyze multi-faceted safety data.

[0004] In traditional safety management processes, the coordination between management, execution, and operational levels is weak. Safety policies formulated by management are prone to information attenuation or execution deviations during downward transmission; on-site data collected by operational levels is also difficult to quickly and accurately relay back to management, preventing management from adjusting strategies in a timely manner based on actual conditions. When a safety incident occurs, the lack of correlation analysis of multimodal data often makes it difficult to quickly trace the source of responsibility or clearly identify the risk propagation path, hindering the optimization and improvement of subsequent safety measures.

[0005] In identifying high-risk work areas, existing methods are mostly based on experience-based judgment or single-indicator assessment, lacking scientific and dynamic analytical basis. This may lead to the neglect of some potentially high-risk areas while overemphasizing some low-risk areas, resulting in a waste of management resources. In terms of safety status assessment and responsibility allocation, there are also problems of strong subjectivity and a lack of quantitative standards, affecting the fairness and effectiveness of safety management. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-level linkage early warning system for construction safety penetration management, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a multi-level linkage early warning system for construction safety penetration management, the system comprising:

[0008] The system comprises a management layer node, an execution layer node, and a job layer node; the management layer node is used to generate a global security policy, the execution layer node is used to decompose security tasks and distribute them to the job layer node, and the job layer node is used to collect on-site security data.

[0009] The system executes the following process:

[0010] Acquire multimodal safety data reported by the operational layer nodes, including equipment vibration characteristics, environmental image streams, personnel positioning trajectories, and safety operation records;

[0011] Based on the changing trend of equipment vibration characteristics over time, stable operating conditions and abnormal operating conditions are divided; high-risk operation areas are screened based on the differences in equipment vibration characteristics between stable and abnormal operating conditions.

[0012] In each high-risk work area, the safety status deviation of each work-level node is calculated by combining the safety task identifiers issued by the execution-level nodes and the environmental image streams reported by the work-level nodes during the stable working period.

[0013] Based on the spatiotemporal correlation between personnel location trajectories and safety operation records during abnormal working conditions, and combined with the safety status deviation, a responsibility traceability weight is generated for each work level node.

[0014] Based on the aforementioned responsibility tracing weights, the root node is selected, and multimodal safety data of the same high-risk work area within the abnormal working condition period are associated to construct a safety risk propagation tree.

[0015] Preferably, the environmental image stream in the multimodal safety data includes timestamps and region identifiers for consecutive frames; the safety operation record includes operation type, operation time, and operator identifier; the system deploys edge computing units at the job layer nodes for feature value extraction of the multimodal safety data.

[0016] Preferably, the edge computing unit performs:

[0017] For the equipment vibration characteristic sequence, calculate the gradient change value of adjacent sampling points, and select the moment when the gradient change value exceeds the preset threshold as the working condition switching point;

[0018] The switching points are arranged in ascending order of gradient change value to generate a switching sequence. The second gradient value of the switching sequence is calculated, and the time corresponding to the maximum second gradient value is used as the dividing point to divide the stable operating condition period and the abnormal operating condition period.

[0019] Compare the average vibration characteristics of the equipment during stable operating periods with those during abnormal operating periods. If the difference in the average values ​​exceeds a preset difference threshold, the corresponding area is marked as a high-risk operating area.

[0020] Preferably, the system further includes a blockchain storage unit, which is used to store a chain of operation records for cross-level interactions;

[0021] The blockchain storage unit executes:

[0022] Receive the security task identifier and corresponding acceptance criteria issued by the execution layer node;

[0023] When the safety operation record of the work layer node triggers the preset key operation type, the current operation time, operator identification and associated environmental image stream feature values ​​are packaged into a data block;

[0024] The data block is verified through consensus based on the aforementioned responsibility traceability weight, and once the verification is successful, it is written into the operation record chain.

[0025] Preferably, the system deploys a cloud-based analytics engine at the management layer node for:

[0026] Extract the feature values ​​of the environmental image stream under the same safety task identifier during the stable working period, and calculate the cosine similarity between the feature values ​​of each operation layer node and the standard feature vector of the execution layer node;

[0027] The safety state deviation is generated by weighting the cosine similarity based on the timestamp interval of consecutive frames in the environmental image stream.

[0028] Preferably, the cloud analytics engine also performs:

[0029] The coordinates of the aggregation point of the personnel location trajectory during the abnormal working condition period are correlated with the operation time of the safety operation record;

[0030] The distribution dispersion of the positioning trajectory within a preset time period before and after the operation time is calculated, and the product of the dispersion and the safety status deviation is normalized to generate the responsibility traceability weight.

[0031] Preferably, the construction of the security risk propagation tree includes:

[0032] Select the job layer node whose responsibility traceability weight exceeds the threshold as the root node;

[0033] Calculate the multimodal safety data similarity between other job layer nodes and the root node, and generate the risk propagation intensity by combining the task urgency weights issued by the execution layer nodes;

[0034] A tree-like hierarchy is constructed in descending order of risk propagation intensity, with nodes at the same level receiving security policy instructions of the same level.

[0035] Preferably, the system deploys a linkage early warning module at the execution layer node, used for:

[0036] Based on the hierarchical structure of the security risk propagation tree, the path from the root node to the first-level child node is defined as a first-level propagation chain;

[0037] Define the longest path from the first-level child node to the leaf node as the second-level propagation chain;

[0038] The first-level early warning strategy is triggered based on the rate of change of real-time environmental image stream feature values ​​of nodes in the first-level transmission chain;

[0039] The secondary early warning strategy is triggered based on the number of sudden changes in the equipment vibration characteristics of nodes in the secondary transmission chain.

[0040] Preferably, the linkage early warning module also performs:

[0041] When the work layer node is offline, the average historical security status deviation and the average historical responsibility tracing weight stored in the edge computing unit are called.

[0042] A lightweight early warning model is generated based on historical averages, and emergency commands are output based on the local matching results of the current equipment vibration characteristics and environmental image stream feature values.

[0043] Preferably, the system further includes a policy feedback unit, used for:

[0044] Collect the execution response times of the primary and secondary early warning strategies;

[0045] The responsibility tracing weight calculation coefficient of nodes in the security risk propagation tree is adjusted according to the response duration.

[0046] The adjusted calculation coefficients are synchronized to the operation record chain of the blockchain storage unit.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] By constructing a three-tiered architecture consisting of management layer nodes, execution layer nodes, and operational layer nodes, the entire process of safety management is seamlessly integrated. The global safety policy generated by the management layer can be accurately decomposed and distributed to the operational layer through the execution layer, ensuring consistency of goals and coordination of actions at all levels. This reduces deviations and attenuation during information transmission, enabling safety management instructions to be implemented efficiently.

[0049] The system integrates and analyzes multimodal safety data reported from the operational level, covering multiple dimensions such as equipment vibration characteristics, environmental image streams, personnel positioning trajectories, and safety operation records, breaking through the limitations of traditional single-data monitoring. By analyzing the trend of equipment vibration characteristics over time, it can scientifically divide stable operating periods into abnormal operating periods, and screen high-risk operating areas based on the difference between the two, making the identification of high-risk areas more objective and dynamic, and helping management resources to be tilted towards areas that truly need attention.

[0050] In high-risk work areas, the safety status deviation of work-level nodes is calculated by combining safety task identifiers with environmental image streams, providing a quantitative basis for assessing the performance of the work-level nodes. This quantitative assessment method reduces the subjectivity of human judgment, making the assessment of safety status more accurate and impartial. Simultaneously, based on the spatiotemporal correlation between personnel location trajectories and safety operation records during abnormal working conditions, and combined with the safety status deviation, responsibility tracing weights are generated, providing data support for responsibility allocation and making the responsibility tracing process clearer and more convincing.

[0051] By selecting root nodes based on responsibility tracing weights and constructing a safety risk propagation tree by associating multimodal safety data from the same high-risk work area during abnormal working conditions, the origin and spread path of risks can be intuitively presented. This helps to quickly clarify the evolution of risks after a safety incident occurs, identify the impact of each link, and provide a clear direction for the adjustment and optimization of subsequent safety measures.

[0052] Through the linkage of three-level nodes and in-depth analysis of multimodal data, the system achieves full-chain management from safety strategy formulation, on-site data collection, high-risk area identification, safety status assessment to responsibility tracing and risk propagation analysis. This penetrating management model strengthens the collaboration between different levels, enhances the overall and systematic nature of safety management, enables the timely detection of potential safety hazards, reduces the probability of safety accidents, and provides a strong analytical foundation for continuous improvement of safety management. Attached Figure Description

[0053] Figure 1 This is a sequence diagram of the multi-level linkage early warning system for construction safety penetration management as described in this invention;

[0054] Figure 2 Flowchart for dividing the operating conditions of edge computing units;

[0055] Figure 3 A flowchart for operating the blockchain storage unit record chain;

[0056] Figure 4 A flowchart for generating the weighting of accountability;

[0057] Figure 5 A flowchart for constructing a security risk propagation tree. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Please see Figure 1 This invention provides a multi-level linkage early warning system for construction safety penetration management. The system includes: a management layer node, an execution layer node, and a work layer node. The management layer node generates a global safety policy, the execution layer node decomposes safety tasks and distributes them to the work layer nodes, and the work layer nodes collect on-site safety data. The system executes the following core processes:

[0060] Acquire multimodal safety data reported by the operational layer nodes, including equipment vibration characteristics, environmental image streams with timestamps and area identifiers, personnel positioning trajectories, and safety operation records containing operation type, operation time, and operator identifiers.

[0061] Based on the changing trend of equipment vibration characteristics over time, stable operating conditions and abnormal operating conditions are divided into periods; by comparing the differences in equipment vibration characteristics between the two periods, high-risk operating areas are screened.

[0062] Within each high-risk work area, the safety status deviation of each work-level node is calculated by combining the safety task identifiers issued by the execution-level nodes with the environmental image streams of the work-level nodes during the stable working period.

[0063] Based on the spatiotemporal correlation between personnel location trajectories and safety operation records during abnormal working conditions, and combined with the safety status deviation, a responsibility traceability weight is generated for each operational level node.

[0064] Root nodes are selected based on responsibility tracing weights, and multimodal safety data of the same high-risk work area during abnormal working conditions are associated to construct a safety risk propagation tree.

[0065] Example 1: See Figure 2 The edge computing units, integrated and deployed at the operational layer nodes, are responsible for the preliminary processing and feature extraction of the multimodal safety data collected on-site. The environmental image streams in the multimodal safety data are captured as video streams, with each frame embedded with precise timestamp information and a unique area identifier code, which corresponds one-to-one with the physical partition of the construction work surface. Safety operation records are entered by operators via mobile terminals or fixed devices; each record includes an operation type classification code, a millisecond-precise operation time record, and an operator identification code. The edge computing units implement differentiated processing procedures for different types of multimodal safety data, focusing on real-time sequence analysis of equipment vibration characteristics.

[0066] The vibration characteristics of the equipment are derived from accelerometers installed on key components of the construction equipment, with a fixed sampling frequency. The edge computing unit first preprocesses the raw vibration signal, including low-pass filtering to eliminate high-frequency noise interference and data normalization to eliminate the influence of different dimensions. After preprocessing, a standardized equipment vibration characteristic sequence is generated. For this sequence, the calculation process is as follows: the numerical difference between each sampling point in the sequence and its next adjacent sampling point is calculated; this numerical difference is defined as the gradient change value of the current point. This gradient change value is continuously monitored and compared with a preset dynamic threshold, which is dynamically generated based on the statistical characteristics of gradient change values ​​within historical sampling windows. When the gradient change value at a certain moment first exceeds the dynamic threshold, that moment is marked as a temporary candidate point for switching operating conditions. The entire sequence is continuously scanned, recording all time points that meet the gradient change value exceeding the limit condition, forming a set of candidate points for switching operating conditions.

[0067] The aforementioned set of candidate points for operating condition switching is arranged in ascending order of gradient change value, forming a gradient switching sequence. A second gradient is calculated for this sequence, and the trend of the second gradient change is analyzed. All points in the sequence are traversed, and the local mean of the data within a set time window before and after each point is calculated. This local mean serves as the local second gradient indicator for that point. Locations where the local second gradient indicator shows abrupt changes or significant increases are identified, and the point with the largest absolute value of the second gradient is determined as the core boundary point. This core boundary point divides the entire equipment vibration characteristic sequence into two parts: the time interval before the core boundary point is determined as the stable operating condition period, and the time interval after the core boundary point is determined as the abnormal operating condition period. This division method relies on the abrupt change characteristics of the data itself and does not require externally preset fixed time parameters.

[0068] After the time periods are divided, the edge computing unit statistically compares the vibration characteristics of the equipment in the two time periods. It calculates the mean vibration characteristics of all sampling points during the stable operating period and the mean vibration characteristics of all sampling points during the abnormal operating period. The absolute difference between these two means is calculated and compared to a preset difference threshold. This preset difference threshold is typically set based on equipment type, historical normal operating data, and engineering experience. If the calculated absolute difference exceeds the preset difference threshold, it indicates a significant abnormal change in the operating conditions of the area where the equipment is located. At this point, the edge computing unit determines the physical location of the equipment based on the area identification code carried in the received environmental image stream frame. This area identification code is marked as a high-risk operation area identifier, and this marking result, the time interval of the time period division, and the vibration characteristic comparison result are transmitted to the upper-level node.

[0069] In environmental image stream processing, the edge computing unit decodes and analyzes the input continuous image frames. Using image processing technology, it extracts key visual features related to construction safety from each frame, such as the wearing status of safety protective equipment by construction workers, the positional contour information of machinery relative to the preset safety area, and the deformation characteristics of temporary support structures. The extraction of these visual features relies on a preset feature recognition model, and the extraction results form a digitized feature vector, which is strictly bound to the timestamp and area identification code of the frame image. Similarly, safety operation records input by operators are structured at the edge, parsing out the operation type code, operation time point, and personnel identification code to form standardized operation event records. All processed local feature data, time period segmentation results, and high-risk area marking results are packaged by the edge computing unit and reported to the execution and management layer nodes via communication links. The entire processing emphasizes localized and real-time data analysis, reducing network transmission burden and improving response speed.

[0070] Example 2: See Figure 3 The blockchain storage unit, as the core management module for cross-level interactive data, constructs a distributed operation record chain structure. After decomposing the safety task, the execution layer nodes encapsulate the safety task identifier and its corresponding acceptance standard specifications into a data packet. This data packet contains the task type code, the task execution area range, the acceptance parameter threshold list, and the expected completion time window information. The above data packet is transmitted by the execution layer nodes to the pending queue of the blockchain storage unit through a dedicated encrypted channel, waiting to be written to the on-chain storage area. At the same time, the operation layer nodes continuously generate safety operation records during the execution of on-site operations. These records are classified and processed within the edge computing unit. The safety operation record data structure contains three basic fields: the operation type code uses a preset global enumeration value (for example, code 101 represents "high-altitude operation protective equipment activated", code 205 represents "heavy machinery lock released"), the operation time is recorded with a timestamp accurate to the millisecond, and the operator identifier is associated with a unique identity code in the enterprise personnel database.

[0071] When a work-layer node detects that the operation type code of a safety operation record matches a preset set of critical operation type codes, it triggers the blockchain data writing process. The set of critical operation type codes is dynamically configured by the management layer node and typically includes operational behaviors involving significant safety risks, such as start / stop commands for high-risk equipment and status changes for safety isolation devices. After identifying a critical operation type code, the edge computing unit performs the following operations: real-time capture of environmental image stream video clips within three seconds before and after the operation; extraction of frame-level feature vectors from the video clips using a built-in convolutional neural network model, with the feature vector dimension fixed at 256 bits; and encapsulation of the operation time, operator identifier, and feature vectors. The encapsulated data structure uses a standard protocol format and includes a header identifier, data body, and checksum. The header identifier is bound to the corresponding safety task identifier and area positioning code.

[0072] The consensus verification process is automatically scheduled by the smart contracts of the blockchain storage units. When a new data block enters the verification queue, the verification process is activated: the smart contract queries the latest generated accountability weight value from the data block's source region (this weight value is calculated and generated by the cloud analytics engine and synchronized within the cluster). The accountability weight value is quantified on a percentage basis and serves as the weight base for node verification participation. The smart contract sets a dynamic verification threshold based on the sum of all weight bases; for example, when the total weight base of participating nodes reaches a set standard, the minimum verification condition is met. The execution layer node, as the initial verification initiator, is allocated a fixed weight base percentage; the remaining verification requests are broadcast to currently online high-weight job layer nodes, and each node calculates its actual voting weight proportionally based on its own accountability weight value. The system sets a five-second time window; when the accumulated voting weight received within the time window exceeds the preset lower limit of the total weight base percentage, the data block is deemed to have passed verification. Data blocks that do not meet this condition are returned to the sending node for reprocessing.

[0073] Verified data blocks enter the on-chain storage stage. The blockchain storage unit adopts a hybrid storage architecture: block header information is recorded on the distributed main chain, including the block hash, timestamp, and parent block pointer; the data body is stored in regional sub-chain nodes, which are deployed according to security task identifiers. Each data block is automatically written with four indexes: timestamp range index, operator identifier index, security task identifier index, and regional location code spatial index. A bidirectional mapping relationship is established between the block header and the data body through content addressing hash; any data tampering will result in abnormal hash value changes. Simultaneously, the smart contract automatically generates a time continuity proof document between the data block and its preceding associated blocks. This proof document includes a timestamp logical sequence verification record and a data feature similarity comparison report. All storage processes update the regional index status table in real time. This table is synchronously pushed to the execution layer nodes for task progress monitoring and is available for global strategy analysis by the management layer nodes. The entire blockchain storage unit's operation record chain construction process achieves time-series traceability, content immutability, and accountability for cross-level operation data.

[0074] Example 3: See Figure 4 The cloud-based analytics engine at the management layer nodes receives pre-processed multimodal safety data from the edge computing units and standard safety task parameters issued by the execution layer nodes. After determining the stable operating period segmentation and high-risk work area markings, the engine performs a quantitative analysis of the status of the work layer nodes within the specified area. The calculation of safety status deviation relies on the comparison process between environmental image stream feature values ​​and preset standards at the execution layer. Environmental image stream feature values ​​originate from compressed feature vectors extracted by the edge computing units, such as the occupancy rate of safety passages in the construction area, the uniformity coefficient of load on aerial work platforms, and the integrity index of protective net structures. The safety task identifiers issued by the execution layer nodes are associated with a set of standard feature vectors, which are jointly generated by the task type and the regional risk level.

[0075] The cloud-based analytics engine performs the following sequence of operations: Extracting the set of environmental image stream feature vectors from all job-layer nodes under the same safety task identifier during stable operating periods. Traversing each node's feature vector in this set and comparing its similarity with the standard feature vectors set by the execution layer. The similarity calculation uses a cosine similarity algorithm, which outputs a similarity score in the range [-1, 1]. Considering the continuous frame characteristics of the video stream, a time decay factor processing mechanism is introduced: The timestamp difference Δt (in seconds) between the current frame and the previous frame is obtained, and a dynamic weight is applied to the similarity score based on the time interval. A larger time interval indicates a lower correlation between the frame data and the current moment, resulting in a higher degree of weight decay. The final safety state deviation is calculated using the following formula:

[0076]

[0077] in: The safety status deviation of the operational layer node; To ensure the number of effective image frames during the stable operating period; Let i be the feature vector of the i-th frame image; The standard feature vectors issued by the execution layer; The time interval between the i-th frame and the (i-1)-th frame; Time decay factor parameter (preset constant value).

[0078] The generation of accountability weights focuses on data from abnormal operating periods. The cloud-based analytics engine first locates the clustered distribution areas of personnel location trajectories within the abnormal period: it slides and scans the coordinates of trajectory points with a 30-second window length, and uses a density clustering algorithm to identify the set of personnel location cluster coordinates. Simultaneously, it analyzes event data marked as critical operation types in the safety operation records, extracting their operation times. And the corresponding location coordinates. Establish a spatiotemporal mapping relationship: for each key operation moment. Search Personnel location trajectory points within the time window ( (The time radius is preset). Calculate the spatial distribution dispersion index of the coordinates of all trajectory points within this time window: use the standard deviation calculation formula to obtain the mean standard deviation of the position coordinates in the XYZ axes, denoted as σ value.

[0079] The weighting of responsibility tracing is synthesized by fusing safety status deviation and spatiotemporal dispersion data. (Taking...) Multiply the value by the σ value and then normalize the product: Scan all work-level nodes within the current high-risk work area. The product values ​​are used, with the maximum value serving as the normalization base. The final accountability weight is determined by the following formula:

[0080]

[0081] in: Assign responsibility tracing weight to the current job level node; The dispersion index of the trajectory points associated with the operation time; This represents the deviation of the node's safety state. This refers to the set of all nodes within the same high-risk work area. Let be the dispersion index of node j; Let be the deviation of node j from its safe state.

[0082] The weight value is constrained within the range of [0,1], and its value reflects both the degree of abnormality in the node's state and the strength of the correlation between personnel gathering behavior. The weight calculation results for accountability are synchronized to the blockchain storage unit in real time to update the data block verification weight benchmark. The final weight value of each node is appended with a timestamp and regional identifier and stored in a distributed database for use by the security risk propagation tree construction module. Normalization processing ensures that the weight data of different regions and different task types are horizontally comparable, providing standardized metrics for upper-level decision-making.

[0083] Example 4: See Figure 5 Taking a bridge jacking construction project as an example, when the system detects that the work area of ​​pier P7 is a high-risk area, it initiates the safety risk propagation tree construction process. This area contains four operational level nodes: jacking equipment control point A, steel beam welding operation point B, hydraulic monitoring station C, and on-site inspection point D. Through preliminary calculations, the responsibility traceability weights for each node are: point A 0.82, point B 0.65, point C 0.41, and point D 0.27. The system sets the root node selection threshold to 0.7, therefore control point A is selected as the root node of the propagation tree.

[0084] The risk propagation intensity calculation revolves around the correlation of multimodal data. Taking the correlation analysis between welding operation point B and control point A as an example: Equipment vibration feature comparison: Extracting the vibration spectrum of the two nodes during the abnormal working condition period, it was found that the vibration waveform of the grinder at point B and the abnormal vibration of the jacking equipment at point A showed a synchronous fluctuation of 0.85 seconds in the 31Hz frequency band; Environmental image stream correlation: Three consecutive frames of images showed that the trajectory of welding slag spatter at point B was heading towards the area of ​​point A, and the image feature similarity reached 72%; Personnel trajectory correlation: Location data showed that the two operators had intersecting paths within five minutes before the working condition switch, with a cumulative intersection time of 120 seconds; The execution layer set the task urgency weight of this area to level 4 (maximum level 5), and the overall propagation intensity value of point B was found to be 0.78.

[0085] Similar methods were used to calculate other nodes: the vibration characteristics of hydraulic monitoring station C and root node A showed weak synchronization (31% similarity), but environmental images showed that the time interval between the abnormal pressure gauge reading at point C and the time when the equipment vibration exceeded the limit at point A was only 1.2 seconds. Personnel trajectory showed that the technician had inspected the equipment at points A and C before the work condition switch. The task urgency weight was set to the default value of level 3, and the propagation intensity was calculated to be 0.55. The multimodal data correlation between patrol point D and point A was weak; only environmental images showed that the patrolman at point D had looked at the area of ​​point A during the abnormal period, and the propagation intensity calculation result was 0.31.

[0086] The system constructs a tree hierarchy in descending order of propagation intensity: Root node layer: Control point A is at the top of the tree, receiving the highest-level instructions. First-level child nodes: Point B, with a propagation intensity > 0.7, becomes a directly associated node of point A, forming the main transmission chain with the root node. Layered merging: Point C, with a propagation intensity of 0.55, is merged into the second-level node layer. Point D, with a propagation intensity of 0.31, is merged into the third-level node layer. Because the monitoring data of point C and point B are linked by equipment (the hydraulic system powers the welding equipment), an auxiliary connection edge is added from C to B.

[0087] The completed safety risk propagation tree has a three-layer topology: First-level transmission chain: A→B path, corresponding to the core operations of steel beam jacking and welding. Second-level transmission chain: B→C→D path, reflecting the extended risks of hydraulic system support inspections. Hierarchical instruction allocation: Point A and its directly associated point B trigger a red alert strategy, including slowing down the jacking equipment and suspending welding operations. Point C receives a yellow alert instruction, requiring double the frequency of hydraulic parameter monitoring. Point D maintains the blue monitoring instruction, only needing to periodically transmit inspection images.

[0088] In practical applications, when similar vibration characteristics are detected again in area P7, the system automatically activates the propagation tree structure: the equipment at control point A and welding point B synchronously enter the protection state within 300 milliseconds, and hydraulic monitoring station C collects pressure fluctuation data every 10 seconds. The image transmission frequency of on-site inspection point D is increased from one frame every ten minutes to one frame per minute, forming a dynamic response network covering the entire area. This tree structure remains unchanged for three consecutive operating cycles until the difference in vibration characteristics of the equipment in the high-risk area falls below the safety threshold and is automatically deactivated.

[0089] Example 5: The linkage early warning module is deployed at the execution layer node and implements graded responses based on the hierarchical structure of the safety risk propagation tree. Continuing with the bridge jacking construction scenario example, after the safety risk propagation tree is constructed, the system defines two types of transmission paths: the first-level transmission chain includes the transmission path from the root node, the jacking equipment control point A, to the directly associated child node, the steel beam welding point B; the second-level transmission chain starts at welding point B, extends to the hydraulic monitoring station C, and terminates at the on-site inspection point D, representing the longest branch path. The system configures differentiated trigger logic and response mechanisms for each level of transmission chain.

[0090] The Level 1 early warning strategy focuses on risks associated with core equipment. During the jacking operation, the module monitors changes in environmental image stream characteristics at nodes along the primary transmission chain in real time. Feature change quantification is achieved through image frame differential technology: for control point A, the contour displacement vector values ​​of key components of the jacking equipment are continuously extracted; for welding point B, the relative distance between the welding torch's working radius and the preset safety boundary is continuously tracked. When the equipment contour displacement rate at control point A exceeds the preset critical parameter, and this over-limit state persists for more than three image frames, the module immediately activates the Level 1 early warning. The on-site response includes three parallel operations: sending an equipment degradation command to the control terminal of root node A; sending a power-off protection signal to the welding equipment relay at child node B; and displaying a 3D positioning alarm model in a pop-up window on the management interface. If welding point B detects that the welding torch has entered the danger radius, even if control point A has not triggered an alarm, the module immediately initiates a local interlocking response, restricts relevant operation permissions, and uploads the abnormal operation trajectory.

[0091] The Level 2 early warning strategy covers the extended risk pathways. For the node sequence of the Level 2 transmission chain, the module periodically counts abrupt changes in equipment vibration characteristics. Vibration sensors deployed at hydraulic monitoring station C output a cumulative value every five minutes: recording the number of times the hydraulic pump vibration amplitude crosses the danger threshold within the monitoring window. If this number exceeds the set upper limit, and simultaneously an abnormal temperature rise is detected in the electrical cabinet at welding point B, the module determines that a conductive equipment fault exists. In this state, a Level 2 early warning is activated, and regional intervention is implemented: a traffic restriction order is broadcast to the area to which all nodes in the Level 2 transmission chain belong; the nearest safety inspection drone is automatically dispatched to the coordinate location; and a maintenance work order is pushed to the emergency team's mobile terminal with a lock on the completion deadline. For end nodes like on-site inspection point D, anomaly detection relies not only on its own equipment data but also incorporates correlation verification—the warning level of point D is only increased when all upstream nodes have generated early warning states.

[0092] The emergency response mechanism is designed for sudden offline situations. During a nighttime construction operation, hydraulic monitoring station C lost connection due to a communication interruption. The system retrieved historical operation records stored in the edge computing unit: the average deviation of the safety status during the last ten normal operations of this node was extracted, which was 0.38, and the average responsibility traceability weight was 0.42. A lightweight decision-making model was constructed using these values: the equipment pressure gauge readings manually entered by the on-site construction worker were converted into a standard feature space, and a feature matching matrix was constructed with the historical average. When the feature vector corresponding to the pressure gauge reading falls into the preset emergency working condition quadrant, the model outputs an equipment protection command. Within five minutes of this command being triggered, the remote center takes over control, forcibly locks the opening of the associated hydraulic valve, and sends an emergency maintenance request. The system continuously polls the offline node status, and immediately corrects the historical average parameters after communication is restored.

[0093] The strategy response feedback forms a closed-loop control. After each early warning strategy is executed, the module accurately records the completion sequence of response actions at each level: the average time from the issuance of the command to the completion of execution by the field equipment for a Level 1 early warning is 1.2 seconds; the average response interval for Level 2 early warnings involving personnel dispatch is 45 seconds. Based on these time records, the system periodically adjusts the key coefficients in the responsibility traceability weight calculation formula: the parameter items associated with response timeout nodes are multiplied by a time penalty factor, and the factor value is reduced proportionally according to the delay ratio. After the parameter adjustment is completed, an update package is distributed to all relevant nodes through the blockchain operation record chain. The update package includes a version verification code and an effective time marker. This mechanism enables the system to automatically optimize the early warning level judgment criteria during continuous operation, adapting to changes in management needs at different construction stages. The module resets the statistical cycle after each parameter upgrade, forming a dynamic iterative system from early warning triggering to response evaluation and then to parameter optimization.

[0094] 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.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-level linkage early warning system for construction safety penetration management, characterized in that, include: Management layer nodes, execution layer nodes, and operational layer nodes; The management layer node is used to generate global security policies, the execution layer node is used to decompose security tasks and distribute them to the operation layer node, and the operation layer node is used to collect on-site security data. The system executes the following process: Acquire multimodal safety data reported by the operational layer nodes, including equipment vibration characteristics, environmental image streams, personnel positioning trajectories, and safety operation records; Based on the changing trend of equipment vibration characteristics over time, the stable operating condition period and the abnormal operating condition period are divided; High-risk work areas are screened based on the difference in equipment vibration characteristics between stable and abnormal operating conditions. In each high-risk work area, the safety status deviation of each work-level node is calculated by combining the safety task identifiers issued by the execution-level nodes and the environmental image streams reported by the work-level nodes during the stable working period. Based on the spatiotemporal correlation between personnel location trajectories and safety operation records during abnormal working conditions, and combined with the safety status deviation, a responsibility traceability weight is generated for each work level node. Based on the aforementioned responsibility tracing weights, the root node is selected, and multimodal safety data of the same high-risk work area within the abnormal working condition period are associated to construct a safety risk propagation tree; The system deploys a cloud-based analytics engine at the management layer node for: Extract the feature values ​​of the environmental image stream under the same safety task identifier during the stable working period, and calculate the cosine similarity between the feature values ​​of each operation layer node and the standard feature vector of the execution layer node; Based on the timestamp intervals of consecutive frames in the environmental image stream, the cosine similarity is weighted to generate a safety state deviation. The cloud analytics engine also performs: The coordinates of the aggregation point of the personnel location trajectory during the abnormal working condition period are correlated with the operation time of the safety operation record; Calculate the spatial distribution dispersion of the point coordinates of the positioning trajectory within a preset time period before and after the operation time, and normalize the product of the spatial distribution dispersion and the safety status deviation to generate the responsibility traceability weight. The mean standard deviation of the position coordinates in the XYZ axes is obtained by using the standard deviation calculation formula and is used as the spatial distribution dispersion.

2. The construction safety penetration management multi-level linkage early warning system according to claim 1, characterized in that, The environmental image stream in the multimodal safety data includes timestamps and region identifiers for consecutive frames; the safety operation record includes operation type, operation time, and operator identifier; the system deploys edge computing units at the job layer nodes for feature value extraction of the multimodal safety data.

3. The construction safety penetration management multi-level linkage early warning system according to claim 2, characterized in that, The edge computing unit performs: For the equipment vibration characteristic sequence, calculate the gradient change value of adjacent sampling points, and select the moment when the gradient change value exceeds the preset threshold as the working condition switching point; The switching points are arranged in ascending order of gradient change value to generate a switching sequence. The second gradient value of the switching sequence is calculated, and the time corresponding to the maximum second gradient value is used as the dividing point to divide the stable operating condition period and the abnormal operating condition period. Compare the average vibration characteristics of the equipment during stable operating periods with those during abnormal operating periods. If the difference in the average values ​​exceeds a preset difference threshold, the corresponding area is marked as a high-risk operating area.

4. The construction safety penetration management multi-level linkage early warning system according to claim 3, characterized in that, It also includes a blockchain storage unit, which is used to store a chain of operation records for cross-level interactions; The blockchain storage unit executes: Receive the security task identifier and corresponding acceptance criteria issued by the execution layer node; When the safety operation record of the work layer node triggers the preset key operation type, the current operation time, operator identification and associated environmental image stream feature values ​​are packaged into a data block; The data block is verified through consensus based on the aforementioned responsibility traceability weight, and once the verification is successful, it is written into the operation record chain.

5. The construction safety penetration management multi-level linkage early warning system according to claim 4, characterized in that, The construction of the security risk propagation tree includes: Select the job layer node whose responsibility traceability weight exceeds the threshold as the root node; Calculate the multimodal safety data similarity between other job layer nodes and the root node, and generate the risk propagation intensity by combining the task urgency weights issued by the execution layer nodes; A tree-like hierarchy is constructed in descending order of risk propagation intensity, with nodes at the same level receiving security policy instructions of the same level.

6. The construction safety penetration management multi-level linkage early warning system according to claim 5, characterized in that, The system deploys a linkage early warning module at the execution layer node for: Based on the hierarchical structure of the security risk propagation tree, the path from the root node to the first-level child node is defined as a first-level propagation chain; Define the longest path from the first-level child node to the leaf node as the second-level propagation chain; The first-level early warning strategy is triggered based on the rate of change of real-time environmental image stream feature values ​​of nodes in the first-level transmission chain; The secondary early warning strategy is triggered based on the number of sudden changes in the equipment vibration characteristics of nodes in the secondary transmission chain.

7. The construction safety penetration management multi-level linkage early warning system according to claim 6, characterized in that, The linked early warning module also performs: When the work layer node is offline, the average historical security status deviation and the average historical responsibility tracing weight stored in the edge computing unit are called. A lightweight early warning model is generated based on historical averages, and emergency commands are output based on the local matching results of the current equipment vibration characteristics and environmental image stream feature values.

8. The construction safety penetration management multi-level linkage early warning system according to claim 7, characterized in that, The system also includes a policy feedback unit, used for: Collect the execution response times of the primary and secondary early warning strategies; The responsibility tracing weight calculation coefficient of nodes in the security risk propagation tree is adjusted according to the response duration. The adjusted calculation coefficients are synchronized to the operation record chain of the blockchain storage unit.

Images