A tower crane structure safety early warning and active regulation method and system

Abstract

The application provides a tower crane structure safety early warning and active regulation method and system, which comprises the following steps: collecting multi-source heterogeneous perception data of the tower crane structure, constructing a multi-dimensional space-time feature vector after time-space synchronization and fusion; inputting the multi-dimensional space-time feature vector into a pre-trained neural network model, outputting a structure state prediction value in a future preset period, the neural network model comprising a first network layer for extracting space correlation features of multi-source data and a second network layer for learning time sequence dependence; based on the structure state prediction value, combining a preset structure safety boundary condition, calculating a dynamic safety index under a future working condition; according to the dynamic safety index and the structure state prediction value, matching a preset multi-level threshold logic, if a preset warning level is triggered, automatically generating and executing a hierarchical regulation instruction. The method realizes accurate prediction of the future state of the tower crane structure and hierarchical active regulation, and guarantees safe and stable operation of the tower crane.

Description

Technical Field

[0001] This invention belongs to the field of construction machinery safety monitoring technology, and specifically relates to a method and system for early warning and active control of tower crane structure safety. Background Technology

[0002] Tower cranes are core equipment in modern high-rise building construction, and their safety is directly related to life, property, and public safety. Currently, tower crane safety mainly relies on regular manual inspections (such as verticality measurements) and overload protection based on torque limiters. The former is infrequent and highly subjective, unable to detect slow-accumulating deformation and sudden dynamic responses; the latter only acts at the moment of overload, which is a "post-event remedy" and cannot warn of progressive risks caused by foundation settlement, structural damage, fatigue accumulation, etc.

[0003] In recent years, the application of information and communication technologies in tower crane safety management has gradually increased, but existing technical solutions generally suffer from problems such as "single-point development" and "low level of intelligence." These solutions mostly focus on real-time monitoring and threshold alarms for single or limited parameters, resulting in relatively isolated functions. For example, Chinese patent CN111422766A proposes a tower crane capable of early warning of tilt and its tilt detection and early warning method, the core of which lies in detecting tower tilt through a tilt sensor and issuing an early warning. Similarly, other technical solutions also largely revolve around the acquisition and display of single or combined parameters such as torque, wind speed, and slewing angle. While these technologies achieve basic monitoring functions, they suffer from two major flaws:

[0004] (1) Lack of predictability: Existing technologies are mostly based on simple threshold judgments of current or historical data, which makes it difficult to provide an effective early warning window in the early stage of accident hazards.

[0005] (2) Lack of closed-loop control: The monitoring system and the tower crane operation control system are isolated from each other, forming an "information island". Even if an alarm is issued, the response still relies entirely on the operator's subjective judgment and manual operation, which poses a risk of human error or delay. The industry regulatory report also clearly points out that some deployed systems "only provide early warnings and do not achieve control effects", failing to play a real control role. Summary of the Invention

[0006] This invention provides a method and system for early warning and active control of tower crane structure safety, which enables accurate prediction and hierarchical active control of the future state of tower crane structure, ensuring the safe and stable operation of tower crane.

[0007] The technical solution of the present invention is as follows:

[0008] A method for early warning and active control of tower crane structural safety includes the following steps:

[0009] S1: Collect multi-source heterogeneous sensing data of the tower crane structure, and construct a multi-dimensional spatiotemporal feature vector after spatiotemporal synchronization and fusion;

[0010] S2: Input the multidimensional spatiotemporal feature vector into a pre-trained neural network model and output the predicted value of the structural state for a future preset time period. The neural network model includes a first network layer for extracting spatial correlation features of multi-source data and a second network layer for learning temporal dependencies.

[0011] S3: Based on the predicted structural state value and combined with the preset structural safety boundary conditions, calculate the dynamic safety index under future working conditions.

[0012] S4: Based on the dynamic safety indicators and the predicted structural state, a preset multi-level threshold logic is matched. If a preset warning level is triggered, a graded control instruction corresponding to the warning level is automatically generated and executed.

[0013] Furthermore, in the aforementioned tower crane structure safety early warning and active control method, in step S1, the multi-source heterogeneous sensing data includes:

[0014] Inclination and displacement data of the tower root collected by tilt and displacement sensors deployed at the corner of the standard section at the bottom of the tower.

[0015] Structural stress data collected by stress sensors installed on the main members of the lattice column;

[0016] The connection preload data is collected by the preload sensor installed at the connection nodes between the lattice column and the foundation, as well as the connection nodes of the standard tower section;

[0017] Load operation parameters obtained in real time from the tower crane safety monitoring system;

[0018] Wind speed, wind direction, and temperature data are obtained from environmental sensors deployed on top of the tower crane.

[0019] Furthermore, in the tower crane structure safety early warning and active control method, the spatiotemporal synchronization and fusion are performed by an edge computing unit. The edge computing unit performs hardware clock synchronization, sliding window value filtering and feature extraction on multi-source heterogeneous sensing data to form a multi-dimensional spatiotemporal feature vector with a unified time reference.

[0020] Furthermore, in the tower crane structure safety early warning and active control method, in step S2, the neural network model is a CNN-LSTM hybrid neural network model, the first network layer is a one-dimensional convolutional neural network layer used to extract spatial correlation features between multi-source data, and the second network layer is a long short-term memory network layer used to capture the temporal inertial features of structural deformation.

[0021] Furthermore, in the aforementioned tower crane structure safety early warning and active control method, the loss function of the CNN-LSTM hybrid neural network model is L=MSE(Y pred Y true )+λ*max(0, θ pred -θ threshold In the formula, MSE is the mean squared error, λ is the penalty coefficient, and θ is the mean squared error. pred To predict the slope, θ threshold The preset tilt rate threshold is used.

[0022] Furthermore, in the aforementioned tower crane structure safety early warning and active control method, in step S3, the dynamic safety index is the dynamic safety factor K, and the calculation formula for the dynamic safety factor K is: K = min(K1, K2); where,

[0023] K1=σ allow / σ max_pred , σ allow σ is the allowable stress of the tower crane structure material. max_pred The predicted maximum stress is calculated by inversion based on the predicted structural state values;

[0024] K2=M resist / M overturning_pred M resist M is the preset anti-overturning moment. overturning_pred The predicted overturning moment is calculated based on the predicted values ​​of the structural state.

[0025] Furthermore, in the aforementioned tower crane structure safety early warning and active control method, in step S4, the graded control instructions include:

[0026] The orange warning command is used to output mandatory warning information to the human-machine interface in the tower crane cab and set soft limits to restrict the tower crane from entering dangerous working conditions.

[0027] The red danger command is used to send a forced intervention command to the safety linkage control box, which is independent of the original tower crane control system, to perform a phased emergency shutdown operation.

[0028] Furthermore, in the aforementioned tower crane structure safety early warning and active control method, the phased emergency shutdown operation executed by the red danger command includes:

[0029] In the first phase, lifting and increasing speed operations are prohibited, while lowering and decreasing speed operations are permitted to perform unloading operations.

[0030] In the second stage, when the structural condition continues to deteriorate as detected by the backup tilt sensor, the slewing brake is released to allow the boom to freely slew back to the position of minimum resistance.

[0031] In the third stage, the main power supply circuit of the tower crane is physically cut off.

[0032] Furthermore, the tower crane structure safety early warning and active control method further includes step S5: based on the spatiotemporal change patterns of the multi-source heterogeneous sensing data and the predicted structural state values, identify the risk cause types and output targeted corrective measures suggestions. The risk cause types include unilateral settlement mode, overall tilt mode and torsion mode.

[0033] A tower crane structure safety early warning and active control system includes:

[0034] Multi-source sensing module is used to collect multi-source heterogeneous sensing data of the tower crane structure;

[0035] An edge processing module, which is connected to a multi-source sensing module, is used to perform spatiotemporal synchronization and fusion of the multi-source heterogeneous sensing data to construct a multi-dimensional spatiotemporal feature vector.

[0036] The intelligent prediction module is used to input the multi-dimensional spatiotemporal feature vector into a pre-trained neural network model and output the predicted value of the structural state for a future preset time period. The neural network model includes a first network layer for extracting spatial correlation features of multi-source data and a second network layer for learning temporal dependencies.

[0037] The safety assessment module is used to calculate the dynamic safety index under future working conditions based on the predicted structural state value and the preset structural safety boundary conditions.

[0038] The hierarchical control module is used to match preset multi-level threshold logic based on the dynamic safety indicators and the predicted structural state values. If a preset warning level is triggered, the module automatically generates and executes the hierarchical control command corresponding to that warning level.

[0039] The beneficial effects of this invention are as follows:

[0040] This invention discloses a method for early warning and active control of tower crane structural safety. By constructing a multi-dimensional spatiotemporal feature vector and combining it with a neural network model composed of a spatial feature extraction layer and a temporal dependency learning layer, it achieves efficient fusion and accurate analysis of multi-source heterogeneous sensing data of tower cranes. This enables proactive prediction of the future state of the structure, and then automatically generates and executes hierarchical control commands based on dynamic safety indicators and multi-level threshold logic. This constructs an autonomous closed-loop control system of perception, prediction, decision-making, and execution, significantly improving the intelligence level and reliability of tower crane operation, effectively reducing the cost of human intervention and safety risks. It can be widely applied to various large-scale construction scenarios and has significant engineering application value and social benefits.

[0041] This tower crane structural safety early warning and active control method optimizes the deployment of multiple types of sensors, including those for tilt angle, displacement, stress, preload, load, and environmental factors, at key mechanical nodes of the tower crane. It also utilizes edge computing units to achieve hardware clock synchronization and feature extraction, constructing a high-dimensional spatiotemporal feature vector. Based on this, a CNN-LSTM hybrid neural network model is employed. One-dimensional convolutional layers automatically learn the spatial correlation features of sensor data from different locations, while long short-term memory network layers capture the temporal inertial features of structural deformation. This achieves high-precision temporal prediction of future structural states, significantly improving the accuracy and foresight of the early warning system.

[0042] This tower crane structural safety early warning and active control method introduces a dynamic safety factor K, dynamically linking the geometric quantities (tilt angle) predicted by AI with the core safety criteria (strength, stability) of the structural engineering, making the safety assessment results closer to the actual engineering situation. Simultaneously, it sets multi-level threshold logic to generate tiered control instructions ranging from "prompt" to "forced intervention." Orange alerts are handled through a human-machine interface for soft control, while red alerts are handled through an independent safety linkage control box to execute phased emergency shutdowns, avoiding secondary risks caused by a one-size-fits-all emergency stop and achieving differentiated and progressive safety intervention.

[0043] This tower crane structure active control system integrates systems through a unified IoT platform and data bus. All data, prediction results, instructions, and execution status are stored in a blockchain-based log, achieving full traceability and tamper-proof operation, meeting safety audit requirements. This technical solution realizes centralized and digital management of tower crane health status, providing technical support for upgrading tower crane safety monitoring from "manual inspection and experience-based judgment" to "intelligent sensing and data-driven" models, and has good potential for widespread application. Attached Figure Description

[0044] Figure 1 This is a flowchart of a tower crane structure safety early warning and active control method according to the present invention;

[0045] Figure 2 This is a schematic diagram of a tower crane structure according to a tower crane structure safety early warning and active control method of the present invention;

[0046] Figure 3 This is an architecture diagram of a tower crane structure safety early warning and active control system according to the present invention;

[0047] In the diagram: 1. Tilt sensor; 2. Displacement sensor; 3. Stress sensor; 4. Preload sensor; 5. Safety monitoring system; 6. Environmental sensor. Detailed Implementation

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The advantages and features of the present invention will become clearer from the following description and claims. It should be noted that the drawings are all in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0049] like Figure 1 As shown, this embodiment provides a method for early warning and active control of tower crane structure safety, which includes the following steps: S1-S4, and may also include S5.

[0050] S1: Collect multi-source heterogeneous sensing data of the tower crane structure, and construct a multi-dimensional spatiotemporal feature vector after spatiotemporal synchronization and fusion.

[0051] like Figure 2 As shown, the multi-source heterogeneous sensing data comprehensively covers the tower crane's structural status, stress conditions, operating parameters, and environmental conditions, specifically including:

[0052] (1) Tower base status data: The data is collected by tilt sensors 1 and displacement sensors 2 deployed at the four corners of the standard section (foundation section) at the bottom of the tower. This location is the final transmission point of all upper loads and overturning moments, and is most sensitive to overall tilt and displacement. The collected data includes the tilt angle and displacement data of the tower base, which can be used to calculate the overall torsional angle and in-plane rigid body displacement of the tower base, directly reflecting the overall tilt state of the tower crane and providing basic data for anti-overturning assessment. Among them, the tilt sensor 1 adopts a high-precision dual-axis MEMS tilt sensor 1 (range ±10°, accuracy ±0.001°), and the displacement sensor 2 adopts a high-performance GNSS positioning module (RTK mode, static accuracy ±2mm).

[0053] (2) Structural stress data: The stress sensors 3 installed at the top and bottom main members of the four lattice columns are used to back-calculate the actual axial force and bending moment of each column, determine whether the load distribution is uniform, whether there is local overload in the structure, and combine with other data to accurately locate the cause of tilting. The stress sensors 3 are fiber optic grating (FBG) strain sensors.

[0054] (3) Connection preload data: The preload data is collected by the preload sensor 4 installed on the key anchor bolts of the lattice column and the concrete foundation, and the high-strength connection bolts of the standard tower section. It serves as a direct indicator of the structural connection safety, and promptly detects early hidden dangers such as loose connection bolts. Its attenuation is an early sign of structural loosening and decreased collaborative performance. The preload sensor 4 adopts an ultrasonic preload monitoring sensor.

[0055] (4) Load operation parameters: These parameters are obtained in real time from the tower crane safety monitoring system 5 ("black box") through standard data interfaces (such as CAN bus, Modbus), including load moment (M), amplitude (R), slewing angle (θ), lifting height (H), etc., reflecting the actual operating conditions of the tower crane and providing load input for risk assessment.

[0056] (5) Environmental data: Wind speed (V), wind direction (α), and temperature (T) data are collected from environmental sensors 6 deployed at the boom end or top of the tower crane to analyze the impact of environmental factors on the structural stability of the tower crane and provide a basis for wind load calculation. Environmental sensor 6 is a three-dimensional ultrasonic anemometer.

[0057] Spatiotemporal synchronization and fusion are performed through the edge computing unit (industrial-grade edge intelligent gateway). The edge computing unit first performs hardware clock synchronization on the above-mentioned multi-source heterogeneous sensing data to ensure that the data collected by different locations and different types of sensors have a unified millisecond-level timestamp, which is the basis for subsequent spatiotemporal fusion analysis. Then, a sliding window midpoint filtering algorithm is used to eliminate instantaneous spike interference, clean the data, and improve data accuracy. Finally, short-term statistical features of the data (such as mean, variance, and rate of change within 10 seconds) are extracted and integrated to form a multi-dimensional spatiotemporal feature vector with a unified time base, reducing the subsequent computational pressure and improving data reliability.

[0058] Meanwhile, the edge computing unit supports 5G, wired fiber optic, and LoRa multimode communication, automatically switches when the main link is interrupted, and has built-in storage that can cache at least 7 days of raw data. It automatically resumes transmission after the network is restored, ensuring data continuity.

[0059] The preprocessed data forms a multidimensional spatiotemporal feature vector Xt=[S t L t E t C t ], where S t L is a structural state vector (including tilt angle, displacement, stress, preload, etc.). t E is the load history subvector. t C is the environmental history sub-vector. t Context encoding for the construction phase.

[0060] S2: Input the multidimensional spatiotemporal feature vector constructed in step S1 into the pre-trained neural network model, and output the predicted value of the structural state for a preset period in the future, so as to realize the forward prediction of the tower crane structural state. Its core goal is not to fit the history, but to learn the dynamic response law of the tower crane structure under multi-source disturbance, so as to extrapolate the future state.

[0061] The neural network model is a CNN-LSTM hybrid neural network model, employing a two-layer architecture of "spatial feature extraction - temporal dependency learning." Combined with digital twin technology, a digital twin model of the tower crane structure is constructed to improve prediction accuracy. Specifically:

[0062] The first network layer is a one-dimensional convolutional neural network (1D-CNN) layer, used to extract spatial correlation features between multi-source data, such as the coordinated change patterns of sensor data at different locations, the correlation between stress and tilt angle, etc., to achieve deep fusion of multi-source data and automatically learn spatial correlation patterns between data at different locations. The second network layer is a long short-term memory network (LSTM) layer, used to learn the temporal inertial features of structural deformation, to capture the changing trend of tower crane structural state over a continuous period of time, such as the cumulative impact of continuous rainfall on foundation settlement, the continuous effect of load changes on structural stress, the impact of temperature changes on structural deformation, etc., to improve long-term prediction accuracy. Its "gating" mechanism enables it to remember long-term dependencies and forget irrelevant information, adapting to the dynamic changes of tower crane structural state.

[0063] To further optimize prediction performance, the loss function of the CNN-LSTM hybrid neural network model is designed as: L=MSE(Y pred Y true )+λ*max(0, θ pred -θ threshold In the formula, MSE is the mean squared error, used to measure the deviation between the predicted and actual values; λ is the penalty coefficient, used to adjust the penalty weight; θ pred To predict the slope, θ threshold By setting a preset tilt rate threshold and adding a penalty term for tilt rates exceeding the threshold, the model is forced to be more sensitive to dangerous trends, thus improving the reliability of predictions. pred The structural state prediction value output by the model, Y true This refers to the actual state values ​​of the structure collected by the sensor.

[0064] The preset future time period can be flexibly set according to actual construction needs, typically ranging from 1 hour to 7 days, combining short-term risk prediction with long-term safety assessment. Simultaneously, the model employs an "initial training + online incremental learning" mechanism. Initial training uses historical project data or high-fidelity simulation data, labeling the actual state Yt+Δt at the future time Δt, and aims to minimize the prediction error (MSE). After system deployment, newly collected data is added to the training pool weekly, fine-tuning the last few layers of the model to continuously adapt it to the unique geological, material, and construction rhythm characteristics of the project, achieving "increasing accuracy with use." Yt+Δt corresponds to Y... true , where is the actual state value of the structure at a certain moment, and is the Y value output by the model. pred(Predicted values) are compared to optimize the model.

[0065] S3: Based on the structural state prediction value output in step S2, and combined with the preset structural safety boundary conditions, calculate the dynamic safety index under future working conditions to quantitatively assess the future safety level of the tower crane structure. This dynamic safety index is directly related to the core safety criteria of structural engineering, ensuring the scientific rigor and practicality of the assessment.

[0066] The dynamic safety index is the dynamic safety factor K, which is calculated using the formula: K = min(K1, K2). K1 and K2 assess structural safety from the two core dimensions of strength and stability, respectively, comprehensively covering the key assessment dimensions of tower crane structural safety. Their specific definitions are as follows:

[0067] Strength safety factor K1: K1 = σ allow / σ max_pred Where σ allow The allowable stress of the tower crane structure material is determined based on the steel type and design specifications of the tower crane structure; σ max_pred The predicted maximum stress is calculated based on the inversion of the predicted structural state values. It is obtained by inversion using a material mechanics and finite element interpolation model, combining the measured data of stress sensor 3 with the predicted tilt angle.

[0068] Stability safety factor K2: K2=M resist / M overturning_pred M resist The preset overturning moment is determined based on the geological report, foundation design drawings, and tower crane model parameters; M overturning_pred The predicted overturning moment is calculated based on the predicted structural state values, and the load eccentricity (e) is calculated by combining the predicted tilt angle. pred ), wind load predicted by environmental sensor 6 (F) wind ) and tower crane operating load (W+F load The comprehensive calculation yields the expression M. overturning_pred =(W+F load )*e pred +F wind *h (where h is the height of the wind load) comprehensively considers the overturning risk under the coupled effect of load and environment.

[0069] The dynamic safety factor K comprehensively considers two core indicators: strength and stability. The smaller the value, the higher the future safety risk of the tower crane structure. This provides a clear quantitative basis for subsequent graded control and enables precise quantitative assessment of risk, which is different from the traditional assessment method based on a single threshold.

[0070] S4: Based on the dynamic safety factor K (dynamic safety index) calculated in step S3 and the predicted structural state output in step S2 (focusing on the predicted tilt rate θ), predThe system matches preset multi-level threshold logic to determine whether a warning level is triggered. If a preset warning level is triggered, it automatically generates and executes a graded control instruction corresponding to that warning level to proactively mitigate risks and ensure the targetedness and operability of the control instructions.

[0071] The multi-level threshold logic corresponds one-to-one with the warning level and the graded control instructions. The core is divided into two key levels: orange warning and red danger, while also covering two auxiliary levels: normal and attention, forming a complete graded system, as detailed below:

[0072] (1) Normal state (blue): when K≥1.8 and θ pred <0.4θ allow (θ) allow When the preset tilt rate threshold is not met, the system only records data and makes periodic reports, without triggering control commands;

[0073] (2) Attention status (yellow): when 1.5≤K<1.8 and 0.4θ allow ≤θ pred <0.7θ allow When this happens, the system will log the changes and send an SMS notification to the safety officer, reminding them to pay attention to the changes in the structural status.

[0074] (3) Warning status (orange): when 1.2≤K<1.5 and 0.7θ allow ≤θ pred <θ allow At the same time, the system outputs mandatory early warning information to the human-machine interface (HMI) in the tower crane cab, including the warning level, the cause of the risk (such as "settlement trend on the northeast side" and "load eccentricity") and safety prompts; at the same time, it sets soft limits to restrict the tower crane from entering dangerous working conditions (such as limiting the lifting height, amplitude and slewing range), temporarily delineates the electronic safety zone, guides the operator to operate in a standardized manner, and avoids the risk from escalating.

[0075] (4) Dangerous state (red): When K < 1.2 and θ pred ≥θ allow At that time, the system sends a forced intervention command to the safety linkage control box, which is independent of the original control system of the tower crane, to perform a phased emergency shutdown operation, so as to maximize the safety of personnel and equipment. This safety linkage control box is the ultimate hardware guarantee for achieving "active immunity".

[0076] The phased emergency shutdown procedure executed under the red danger command balances safety and rationality, avoiding secondary risks caused by sudden shutdown. It specifically includes three phases:

[0077] Phase 1 (0-15 seconds): Warning and preparation. Trigger a high-decibel audible and visual alarm on site, and simultaneously send a "Do not lift, do not increase" block signal to the tower crane's main control system; at this time, the operator can still perform "lower" and "retract boom" operations to unload the load and quickly reduce the stress on the tower crane structure.

[0078] Phase 2 (15-30 seconds): Passive risk avoidance. If the structural condition continues to deteriorate as detected by the backup tilt sensor 1 (independent of the main sensor, improving reliability), the slewing brake is released to allow the boom to freely rotate to the position of minimum resistance, reducing the stress on the structure and achieving a passive safety state design.

[0079] Phase 3 (30 seconds): Final disconnection. The main power supply circuit of the tower crane is physically cut off by a high-capacity relay in the safety linkage control box, completely stopping the operation of the tower crane. At the same time, the "executed" status signal is fed back to the cloud monitoring platform to notify the staff to handle the situation on site.

[0080] The safety linkage control box adopts dual PLC hot standby, independent battery power supply (72-hour battery life), manual emergency override button, and shockproof and splashproof shell to ensure its absolute reliability under extreme working conditions. At the same time, when receiving red danger commands, it will perform digital signature verification and timestamp freshness verification to prevent forgery or replay attacks and ensure the security of command execution.

[0081] S5: Based on the spatiotemporal variation patterns of the multi-source heterogeneous sensing data collected in step S1 and the structural state prediction values ​​output in step S2, identify the types of risk causes and output targeted corrective measures to help staff resolve risks at their source, enrich the beneficial effects of the technology, and make the solution more practical and complete.

[0082] The risk causes are mainly categorized into three core modes: unilateral settlement mode, characterized by a continuous increase in tower tilt on one side and uneven displacement, often caused by uneven foundation settlement; overall tilt mode, characterized by overall tower tilt with consistent displacement trends at each corner, often caused by long-term load eccentricity or accumulated environmental factors; and torsion mode, characterized by excessive tower torsion angle and uneven stress distribution in different directions, often caused by uneven preload of connecting bolts or torsional load action. Corresponding corrective suggestions are provided for each mode. For example, for unilateral settlement mode, it is recommended to check the foundation settlement and reinforce it; for torsion mode, it is recommended to check the preload of connecting bolts and adjust the load distribution; and for overall tilt mode, it is recommended to adjust the load distribution and check the structural integrity of the tower. This step provides on-site safety management personnel with guidance on risk root cause analysis and handling, improving the system's diagnostic capabilities and engineering practicality, and addressing the technical deficiency of existing systems that "only alarm, but do not diagnose."

[0083] The above method, by constructing multi-dimensional spatiotemporal feature vectors and combining them with a neural network model composed of a spatial feature extraction layer and a temporal dependency learning layer, achieves efficient fusion and accurate analysis of multi-source heterogeneous sensing data of tower cranes. This enables proactive prediction of the future state of the structure, and then automatically generates and executes hierarchical control commands based on dynamic safety indicators and multi-level threshold logic. This constructs an autonomous closed-loop control system of perception, prediction, decision-making, and execution, significantly improving the intelligence level and reliability of tower crane operation, effectively reducing the cost of human intervention and safety risks. It can be widely applied to various large-scale construction scenarios and has significant engineering application value and social benefits.

[0084] like Figure 3 As shown, this embodiment also provides a tower crane structure safety early warning and active control system, including a multi-source sensing module, an edge processing module, an intelligent prediction module, a safety assessment module, and a hierarchical control module. These modules are integrated through a unified IoT platform and data bus, working collaboratively to form a fully closed-loop active control architecture encompassing intelligent sensing, fusion prediction, dynamic assessment, and decision-making control, ensuring seamless integration of all aspects. Details are as follows:

[0085] The multi-source sensing module is used to collect multi-source heterogeneous sensing data of the tower crane structure. This module consists of various types of sensors, optimized based on the analysis results of the tower crane structural mechanics model, covering key mechanical paths. These include high-precision dual-axis MEMS tilt sensors 1 and high-performance GNSS displacement sensors 2 deployed at the four corners of the standard section (foundation section) at the bottom of the tower body; fiber optic grating (FBG) strain sensors at the top and bottom main members of the four lattice columns; ultrasonic preload monitoring sensors at the connection nodes between the lattice columns and the foundation, and at the connection nodes of the standard sections of the tower body; and a three-dimensional ultrasonic anemometer at the end of the tower crane boom or the top of the tower. Simultaneously, it connects to the tower crane safety monitoring system 5 ("black box") via standard data interfaces (such as CAN bus and Modbus) to acquire real-time operating parameters such as load torque and amplitude, achieving multi-dimensional and comprehensive data acquisition. This ensures that the data covers the core influencing factors of tower crane structural safety, providing high-quality raw data for subsequent processing and prediction.

[0086] The edge processing module connects to the multi-source sensing module to perform spatiotemporal synchronization and fusion of multi-source heterogeneous sensing data, constructing a multi-dimensional spatiotemporal feature vector. This module employs an industrial-grade edge computing unit (edge ​​smart gateway), incorporating a hardware clock synchronization module, a sliding window mid-range filtering data cleaning algorithm, and a short-time statistical feature extraction algorithm. It achieves millisecond-level synchronization of multi-source data, instantaneous interference elimination, and feature integration, outputting a high-quality multi-dimensional spatiotemporal feature vector to provide data input for the intelligent prediction module. Simultaneously, the edge computing unit supports 5G, wired fiber optic, and LoRa multi-mode redundant communication, featuring data interruption resume and at least 7 days of data caching capabilities. This reduces cloud computing pressure, improves system response speed and data continuity, and ensures no data loss even in extreme conditions.

[0087] The intelligent prediction module is used to input multi-dimensional spatiotemporal feature vectors into a pre-trained neural network model and output predicted structural states for a preset future time period. This module incorporates a CNN-LSTM hybrid neural network model, including a one-dimensional convolutional neural network layer (the first network layer, used to extract spatial correlation features) and a long short-term memory network layer (the second network layer, used to capture temporal dependent features). Combined with digital twin technology, it constructs a digital twin model of the tower crane structure, improving prediction accuracy. Simultaneously, it integrates a model training and online optimization module, which can continuously optimize model parameters based on historical and real-time data, achieving "initial training + online incremental learning" to further improve prediction accuracy. The model loss function adopts L=MSE(Y pred Y true )+λ*max(0, θ pred -θ threshold This ensures sensitivity to dangerous trends, optimizes tilt rate prediction in a targeted manner, and improves model robustness.

[0088] The safety assessment module calculates dynamic safety indicators under future operating conditions based on predicted structural state values ​​and preset structural safety boundary conditions. This module has a built-in dynamic safety factor calculation model that automatically calls preset parameters such as allowable material stress and overturning moment to accurately calculate the strength safety factor K1 and stability safety factor K2, ultimately outputting the dynamic safety factor K=min(K1, K2). It also stores preset multi-level threshold logic (covering four levels: normal, attention, warning, and danger), providing clear decision-making basis for the graded control module, achieving quantitative risk assessment, and ensuring that the assessment results are linked to structural engineering design specifications, possessing scientific rigor and authority.

[0089] The tiered control module matches preset multi-level threshold logic based on dynamic safety indicators and structural status predictions. If a preset warning level is triggered, it automatically generates and executes the corresponding tiered control command. This module includes a human-machine interface (HMI) unit and a safety linkage control unit. The HMI unit (in-cab HMI screen) outputs information related to orange warnings and yellow alerts, along with soft limit control commands. During a warning, the current operating interface is highlighted with a prominent color, and a safe working envelope is graphically displayed. The operator must click "Confirm" to temporarily close the pop-up window, ensuring information delivery. The safety linkage control unit is completely independent of the original tower crane control system. It receives red danger commands and executes phased emergency shutdown operations. It features dual PLC hot standby, independent power supply, and manual emergency operation, ensuring reliable operation under extreme conditions. It can also integrate a risk cause identification module to analyze the risk causes in step S5 and output corrective action suggestions, helping staff mitigate risks at their source.

[0090] This system, through the collaborative work of multi-source sensing modules, edge processing modules, intelligent prediction modules, safety assessment modules, and hierarchical control modules, forms a complete safety closed loop of early risk perception, intelligent decision generation, and automatic action execution. Compared with existing technologies, the system of this invention possesses the capabilities of self-sensing, self-predicting, self-decision-making, and self-execution, realizing a paradigm shift in tower crane structural safety from passive response to proactive prevention and control, and effectively solving the technical defects of existing systems such as delayed early warning and passive control.

[0091] The implementation of the technical solution of the present invention will be described in detail in a non-limiting manner through two embodiments.

[0092] Example 1: Long-term health management of climbing luffing tower cranes in super high-rise buildings

[0093] This embodiment is applied to a climbing luffing tower crane (model: M1280D) inside the core tube of a 350-meter super high-rise building, focusing on demonstrating the system's ability to predict and preventively maintain long-term, gradually changing structural risks.

[0094] (a) Scenario-specific system configuration

[0095] 1. Sensor selection and installation:

[0096] Inclination sensor 1: The TILT-3D model from SpectraPrecision, Switzerland, is selected. It has a measurement range of ±15°, a static accuracy of ±0.001°, and built-in temperature compensation. It is rigidly connected to the upper part of the tower foundation section and the connecting flange of the first standard section at four diagonal positions via custom clamps. These positions, determined by finite element analysis, are the critical sections for the transmission of overall bending moment to the foundation, and can most sensitively reflect the overall tilt state of the tower crane.

[0097] Stress sensor 3: A fiber optic grating (FBG) strain sensor, model S-FBG-10, from HBM, Germany, with a measuring range of ±1500µm / m, is used. A total of 32 measuring points are symmetrically attached to the main members of the four lattice columns, 1 meter below the top foundation plate and 2 meters above the bottom interface with the rock strata. Before installation, the bonding surfaces are sandblasted to increase adhesion strength, and permanent encapsulation with epoxy resin and a metal shield is performed to ensure the long-term stability of the sensor in harsh construction environments.

[0098] Bolt preload sensor 4: Among the 32 M42 high-strength anchor bolts connecting the lattice column and the concrete foundation, 4 bolts at the diagonal position are selected and replaced with Piezotronics series wireless preload smart washers from PCB Company of the United States to directly measure the axial force of the bolts with an accuracy of ±1.5%FS.

[0099] 2. Data Networks and Edge Computing

[0100] Industrial-grade wireless APs are installed every 150 meters within the core to build a 5G millimeter-wave private network, ensuring data uplink latency of less than 20ms and bandwidth of more than 100Mbps to meet the requirements for real-time high-frequency data transmission.

[0101] The edge gateway uses Huawei Atlas500 intelligent mini-stations, with a built-in ARM architecture processor capable of running lightweight data cleaning algorithms. This gateway is responsible for synchronizing the hardware clock for 32 strain signals, 4 tilt signals, 4 preload signals, and multi-source data such as load and environmental data. It also eliminates instantaneous spike interference through sliding window mid-range filtering, extracts statistical features such as mean, variance, and rate of change over 10 seconds, and then packages and uploads these features to the cloud.

[0102] (II) Detailed process of model initialization and training

[0103] After the tower crane is installed and commissioned, the system will undergo a 90-day model initialization training and verification period.

[0104] 1. Data Acquisition Plan

[0105] Month 1 (Establishment of Static Load Baseline): Collect static load response data of the tower crane at different attachment heights (every two climbs). Conduct a static load test at 75% of rated load once a week, record the baseline values ​​of stress and tilt angle after stabilization, and establish an initial mechanical characteristic library of the tower crane structure under different attachment states.

[0106] The second month (dynamic response learning): Under controllable conditions, simulate dynamic load conditions, such as sudden unloading (simulating unhooking of the load) and slow rotation (simulating normal operation), and collect transient response data of the structure to enable the model to learn the deformation response characteristics of the tower crane under dynamic loads.

[0107] Third month: Based on weather forecasts, focus on collecting monitoring data under wind conditions of level 6 (10.8-13.8 m / s) to establish the mapping relationship between wind load and structural response.

[0108] 2. Model Building and Training

[0109] An improved “CNN-BiLSTM-Attention” hybrid neural network model is adopted, which adds bidirectional LSTM and attention mechanism to improve long-term prediction accuracy compared to the basic CNN-LSTM hybrid neural network model.

[0110] Model input: Time series data window of the past 30 days, sampling interval of 10 minutes, 45 feature dimensions (including 32 strain features, 4 tilt angle features, 4 preload features, 2 load features, 2 environmental features, and 1 construction stage code).

[0111] The CNN component employs two one-dimensional convolutional layers. The first layer has 64 kernels of size 3, and the second layer has 32 kernels of size 3. These layers are specifically designed to extract the spatial relationship between the stress of the four lattice columns and the overall tilt. For example, the convolutional kernels can automatically learn the pattern of "when the stress of column 1 increases and the stress of column 3 decreases, in which direction is the whole structure more likely to tilt?"

[0112] The BiLSTM part employs a bidirectional long short-term memory network with 128 units per layer. This bidirectional structure allows the model to utilize both past and future contextual information, making it particularly effective for identifying the cumulative effects of creep deformation. For example, the model can learn a long-term dependency: "Several consecutive days of rainfall lead to an increase in soil moisture content, which in turn causes a slow increase in the tilt rate over the next few days."

[0113] Attention mechanism: An attention layer is introduced after the BiLSTM layer, enabling the model to focus on data segments after drastic load changes or strong winds. These periods often contain key features of structural response, and the prediction accuracy is improved by assigning them higher weights.

[0114] Training configuration: The first 75 days of data were used as the training set, and the last 15 days as the validation set. The Adam optimizer was used with an initial learning rate of 0.001, and the ReduceLROnPlateau strategy was employed to automatically reduce the learning rate during the plateau phase of the loss function.

[0115] The loss function is: Loss = MSE(θ) pred θ true )+0.5×MAE(σ pred ,σ true )+0.1×max(0,θ pred-0.006)^2. The first term ensures the accuracy of the tilt angle prediction, the second term takes into account the accuracy of the stress prediction, and the third term penalizes cases where the predicted tilt rate exceeds 0.006 rad (approximately 0.34°), making the model more sensitive to dangerous trends.

[0116] The model converged after training for approximately 300 epochs. On the validation set, the mean absolute error (MAE) of the predicted tilt angle for the next 7 days was less than 0.0003 rad, indicating that the model has good long-term predictive ability.

[0117] (III) In-depth technical analysis of early warning examples

[0118] When construction reaches a height of 180 meters (after the Nth layer of attachment), the system triggers an orange alert. The internal technical process is as follows:

[0119] 1. Data anomaly detection

[0120] The system's backend change point detection algorithm (using the PELT algorithm, i.e., the PrunedExactLinearTime algorithm) first identified the stress reading of the FBG sensor (numbered FBG-3B) at the bottom of column 3. This reading showed a slow, monotonous increase exceeding three standard deviations over 24 consecutive hours, while the load did not increase accordingly. This anomaly is easily overlooked in traditional fixed-threshold alarm systems because the absolute stress value remains within a safe range. However, this invention detected the potential risk in advance through trend analysis.

[0121] 2. Model Prediction and Cross-Validation

[0122] The AI ​​prediction model, based on data from the previous week that included this anomaly, outputs the state for the next 7 days. The prediction shows that the northeast corner tilt angle α will increase from the current 0.0021 rad to 0.0036 rad (an increase of 71%). Simultaneously, a parallel verification module based on a simplified mechanical model, separately estimating the tilt increment based on the stress increment of column 3, matches the AI ​​prediction trend, forming cross-validation and ruling out the possibility of false alarms due to a single point of sensor failure. This dual verification mechanism of "AI prediction + physical model verification" significantly improves the reliability of the early warning.

[0123] 3. Calculation of dynamic safety factor

[0124] Input parameters:

[0125] Predicted tilt angle: α pred =0.0036rad, β pred =0.0015 rad

[0126] Maximum wind speed forecast for the next week: V f =12m / s

[0127] Current maximum planned load: M plan =3200kN·m

[0128] Calculation process:

[0129] According to the "M1280D Tower Crane Foundation Design Calculation Sheet", the design value of the foundation overturning moment M under the current attachment condition is found. resist =8500 kN·m. Calculate and predict the overturning moment M. overturning_pred =(Structural self-weight eccentricity) + (lifting eccentricity) + (wind load moment) ≈ 14500kN·m.

[0130] Stability safety factor: K stability =18500 / 14500≈1.28. Meanwhile, the strength safety factor K is calculated based on the predicted stress. strength =1.65. Dynamic safety factor: K=min(1.28,1.65)=1.28.

[0131] 4. Early warning triggering and decision support

[0132] Since K=1.28<1.5, the system triggered an orange alert. The system not only displayed a warning to the HMI in the driver's cab, but its root cause analysis module automatically correlated the abnormal stress and preload data of column No. 3 (showing that the preload of one anchor bolt in this column had decreased by 15%). The alarm message clearly stated: "High risk: Accelerated tilting on the northeast side, directly related to suspected loosening of the anchor bolts connecting the foundation of the 3-lattice column. Recommended priority: Immediate inspection and tightening."

[0133] 5. Closed-loop handling and effect feedback: Following system instructions, maintenance personnel used a hydraulic wrench to retighten the four anchor bolts of column No. 3 to the design preload (625 kN). After the handling, the system monitored that the stress in column No. 3 returned to the normal range within 4 hours, the tilt angle stopped increasing and a slight rebound occurred.

[0134] The incremental learning module automatically stores this entire "anomaly-handling-recovery" cycle of data as a high-quality sample in the knowledge base. This sample includes anomaly feature patterns (monotonically increasing stress, decreasing preload), intervention measures (anchor bolt tightening), and response effects (stress reduction), used to optimize the recognition accuracy of similar patterns in the future. The system truly achieves "better accuracy with use."

[0135] This embodiment focuses on long-term health management, demonstrating how the system, through sophisticated sensor deployment, ample data collection and training, and the long-term predictive capabilities of deep learning models, enables early identification and preventative maintenance of risks to slowly changing structures. Its core value lies in "early detection, accurate diagnosis, and timely intervention," eliminating safety risks in their nascent stage.

[0136] Example 2: Real-time emergency support for tower cranes next to deep foundation pits in complex urban areas

[0137] This embodiment is applied to a QTZ80 tower crane next to a deep foundation pit in a subway, highlighting the system's ability to instantly perceive, quickly predict, and intervene in sudden and rapidly changing risks.

[0138] (a) Rapid deployment and model hot start

[0139] Since the system was installed using a tower crane, it had to be put online in a very short time.

[0140] Quick-installation kit: All sensors use high-strength magnetic bases with universal locking mechanisms, allowing installation and coarse adjustment of a measurement point to be completed within 20 minutes. The GNSS antenna uses a quick-release clamp bracket.

[0141] Wireless self-organizing network: Adopting a mesh network based on the LoRa protocol, sensor nodes and relay gateways automatically find their way to each other without the need for complex wiring. A single gateway can cover a radius of 500 meters, meeting the needs of most construction sites.

[0142] Model warm start (transfer learning):

[0143] ① Retrieve a general model trained in a similar soft soil geological condition project from the cloud model library as the basis.

[0144] ② Using the monitoring data from the first 48 hours after installation (even if the data volume is small), the last two layers (fully connected layers) of the model were quickly fine-tuned. A relatively large initial learning rate (0.01) and a small number of epochs (about 50 rounds) were used to enable the model to quickly adapt to the "personal" characteristics of this tower crane, including differences in foundation stiffness, sensor installation deviations, and surrounding environmental characteristics.

[0145] ③ After fine-tuning, the system enters the "monitoring-early warning" state, while the background continues to collect data to prepare for the formal incremental training each week. The entire hot start process was completed within 72 hours after installation, achieving rapid deployment and launch of the system.

[0146] (II) Millisecond-level response chain analysis of emergency cases

[0147] One day, a sudden water seepage emergency occurred in the foundation pit, and the structural condition of the tower crane deteriorated rapidly. The system response was a highly automated assembly line:

[0148] (1) High-frequency data acquisition (t=0s)

[0149] The edge gateway's built-in emergency mode automatic trigger condition detected that the tilt angle change rate exceeded 0.0001 rad / s within a 10-second window. The system immediately and automatically switched the data acquisition frequency from the normal 1Hz to the emergency mode 10Hz to capture a more detailed deformation process.

[0150] The data stream from tilt sensor 1 shows that the tilt angle α in the northeast direction increased linearly from 0.0018 rad to 0.0035 rad between t0 and t0+120s, a rate of change far exceeding that under normal operating conditions.

[0151] (2) Rapid prediction and iterative update

[0152] Initial prediction (t0+10s): Based on the acceleration trend in the first 10 seconds, the model initially predicts that the limit may be exceeded in the next 5 minutes. A yellow alert is triggered, and the safety officer is notified to pay attention.

[0153] Continuous prediction (t0+30s, 60s, 90s): The model re-predicts every 10 seconds using a rolling time window. As the measured tilt accelerates, the predicted overshoot time decreases from 5 minutes to 2 minutes. The system continuously outputs prediction results, and the dynamic safety factor K decreases over time.

[0154] Finally, at t0+90s, the model predicted that the tilt would exceed the allowable value (0.004 rad) after 10 minutes. At this point, the dynamic safety factor K, calculated based on the real-time stress and the predicted tilt angle, had decreased to 1.15.

[0155] (3) Red danger trigger and hard intervention command generation (t0+95s)

[0156] The system core controller determines that K = 1.15 < 1.2, triggering a red danger alarm. It automatically generates a "Red Alert - Emergency Shutdown" command package. This command package is in JSON format, contains the command type, timestamp, and unique sequence number, and is encrypted using a digital signature based on the RSA algorithm.

[0157] (4) Deterministic response of the safety linkage control box

[0158] The safety linkage control box is independent of the original tower crane control system. It uses a Siemens S7-1500 series PLC as the core controller and has the following response process:

[0159] Command reception and verification (t0+95s to t0+96.5s): The control box receives commands simultaneously through dual communication channels (5G private network + 4G backup) to ensure reliable command delivery. The main control PLC first verifies the digital signature and timestamp (verifying that the command was issued within 5 seconds). After confirming that the command is legal and fresh (within 5 seconds), it enters the execution preparation state.

[0160] Phase 1 Execution - Functional Suppression (t0+96.5s to t0+97.0s): The PLC immediately sets two safety relays. The first relay disconnects, outputting a dry contact disconnect signal to the "Lift Up" and "Amplitude" control circuits of the original tower crane control system, physically prohibiting dangerous actions (continued lifting and outward luffing). The second relay closes, triggering a 120-decibel rotating strobe light and siren on site.

[0161] Monitoring and Judgment Window (t0+97.0s to t0+112.0s): The system continuously reads data from the backup tilt sensor 1, which is directly connected to the control box. During this period, the operator uses the still operable "lower" and "reduce" functions to unload the load to the ground within 15 seconds.

[0162] Phase Two - Passive Risk Avoidance (t0+112.0s): Since the tilt angle remained unstable after unloading, it continued to increase slightly to 0.0036 rad. The PLC determined that the structural condition was still deteriorating and triggered the third relay, cutting off the power supply to the slewing brake. After the slewing brake was released, the boom began to slowly and freely rotate under wind load, automatically adjusting to the downwind direction, effectively reducing the lateral wind-exposed area.

[0163] Final Execution - Energy Isolation (t0+113.0s): One second after the second stage execution (i.e., t0+113.0s), the PLC drives the 400A high-capacity main contactor to completely cut off the tower crane's main power supply circuit. At this point, the tower crane enters a complete shutdown state, and all power sources are physically isolated.

[0164] The control box will contain timestamps of all actions and confirmation messages of the final status, which will be sent back to the cloud via a dedicated 4G network for subsequent traceability and analysis.

[0165] (5) Post-event analysis and tracing: After the danger is eliminated, the system automatically generates an event report. Key data include: tilt change time series curve (sampling rate 10Hz), comparison of predicted values ​​and actual values ​​at each stage, timestamps of each action of the safety linkage control box and dynamic safety coefficient K value at the time of warning trigger.

[0166] The system aligned the tilting abrupt change data with the sudden increase in pore water pressure data from the foundation pit monitoring system on the timeline, clearly revealing the causal chain of "water seepage → soil softening → decrease in foundation bearing capacity → tower crane tilting". This complete chain of data evidence provides irrefutable objective evidence for determining responsibility and formulating subsequent reinforcement plans, effectively avoiding disputes among the parties involved.

[0167] This embodiment focuses on real-time emergency response, demonstrating how the system achieves instantaneous perception and mandatory intervention in sudden risks through high-frequency data acquisition, rapid predictive iteration, and deterministic response from an independent safety control box. Its core value lies in "rapid response, tiered execution, and reliable protection," ensuring the safety of equipment and personnel under extreme operating conditions.

[0168] The above description is merely a description of preferred embodiments of the present invention and is not intended to limit the scope of the present invention in any way. Any changes or modifications made by those skilled in the art based on the above disclosure shall fall within the protection scope of the claims.

Claims

1. A method for early warning and active control of tower crane structural safety, characterized in that, Includes the following steps: S1: Collect multi-source heterogeneous sensing data of the tower crane structure, and construct a multi-dimensional spatiotemporal feature vector after spatiotemporal synchronization and fusion; S2: Input the multidimensional spatiotemporal feature vector into a pre-trained neural network model and output the predicted value of the structural state for a future preset time period. The neural network model includes a first network layer for extracting spatial correlation features of multi-source data and a second network layer for learning temporal dependencies. S3: Based on the predicted structural state value and combined with the preset structural safety boundary conditions, calculate the dynamic safety index under future working conditions. S4: Based on the dynamic safety indicators and the predicted structural state, a preset multi-level threshold logic is matched. If a preset warning level is triggered, a graded control instruction corresponding to the warning level is automatically generated and executed.

2. The tower crane structure safety early warning and active control method as described in claim 1, characterized in that, In step S1, the multi-source heterogeneous sensing data includes: Inclination sensor (1) and displacement sensor (2) deployed at the corner of the standard section at the bottom of the tower body collect the tilt angle and displacement data of the tower root; Structural stress data collected by stress sensors (3) installed on the main members of the lattice column; The connection preload data is collected by the preload sensor (4) installed at the connection nodes of the lattice column and the foundation, as well as the connection nodes of the standard tower section; Load operation parameters obtained in real time from the tower crane safety monitoring system (5); Wind speed, wind direction and temperature data obtained from environmental sensors (6) deployed on top of the tower crane.

3. The tower crane structure safety early warning and active control method as described in claim 1, characterized in that, The spatiotemporal synchronization and fusion are performed through an edge computing unit, which performs hardware clock synchronization, sliding window value filtering and feature extraction on multi-source heterogeneous sensing data to form a multi-dimensional spatiotemporal feature vector with a unified time reference.

4. The tower crane structure safety early warning and active control method as described in claim 1, characterized in that, In step S2, the neural network model is a CNN-LSTM hybrid neural network model. The first network layer is a one-dimensional convolutional neural network layer, which is used to extract spatial correlation features between multi-source data. The second network layer is a long short-term memory network layer, which is used to capture the temporal inertial features of structural deformation.

5. The tower crane structure safety early warning and active control method as described in claim 4, characterized in that, The loss function of the CNN-LSTM hybrid neural network model is L=MSE(Y). pred Y true )+λ*max(0, θ pred -θ threshold In the formula, MSE is the mean squared error, λ is the penalty coefficient, and θ is the mean squared error. pred To predict the slope, θ threshold The preset tilt rate threshold is used.

6. The tower crane structure safety early warning and active control method as described in claim 1, characterized in that, In step S3, the dynamic safety index is the dynamic safety coefficient K, and the formula for calculating the dynamic safety coefficient K is: K = min(K1, K2); where, K1=σ allow / σ max_pred , σ allow σ is the allowable stress of the tower crane structure material. max_pred The predicted maximum stress is calculated by inversion based on the predicted structural state values; K2=M resist / M overturning_pred M resist M is the preset anti-overturning moment. overturning_pred The predicted overturning moment is calculated based on the predicted values ​​of the structural state.

7. The tower crane structure safety early warning and active control method as described in claim 1, characterized in that, In step S4, the graded control command includes: The orange warning command is used to output mandatory warning information to the human-machine interface in the tower crane cab and set soft limits to restrict the tower crane from entering dangerous working conditions. The red danger command is used to send a forced intervention command to the safety linkage control box, which is independent of the original tower crane control system, to perform a phased emergency shutdown operation.

8. The tower crane structure safety early warning and active control method as described in claim 7, characterized in that, The phased emergency shutdown operation executed by the red danger command includes: In the first phase, lifting and increasing speed operations are prohibited, while lowering and decreasing speed operations are permitted to perform unloading operations. In the second stage, when the structural condition is continuously deteriorating as detected by the backup tilt sensor (1), the slewing brake is released to allow the boom to freely slew to the position of minimum resistance. In the third stage, the main power supply circuit of the tower crane is physically cut off.

9. The tower crane structure safety early warning and active control method as described in claim 1, characterized in that, It also includes step S5: based on the spatiotemporal change patterns of the multi-source heterogeneous sensing data and the predicted structural state values, identify the risk cause types and output targeted corrective measures suggestions. The risk cause types include unilateral settlement patterns, overall tilting patterns and torsion patterns.

10. A tower crane structural safety early warning and active control system, characterized in that, include: Multi-source sensing module is used to collect multi-source heterogeneous sensing data of the tower crane structure; An edge processing module, which is connected to a multi-source sensing module, is used to perform spatiotemporal synchronization and fusion of the multi-source heterogeneous sensing data to construct a multi-dimensional spatiotemporal feature vector. The intelligent prediction module is used to input the multi-dimensional spatiotemporal feature vector into a pre-trained neural network model and output the predicted value of the structural state for a future preset time period. The neural network model includes a first network layer for extracting spatial correlation features of multi-source data and a second network layer for learning temporal dependencies. The safety assessment module is used to calculate the dynamic safety index under future working conditions based on the predicted structural state value and the preset structural safety boundary conditions. The hierarchical control module is used to match preset multi-level threshold logic based on the dynamic safety indicators and the predicted structural state values. If a preset warning level is triggered, the module automatically generates and executes the hierarchical control command corresponding to that warning level.

Images