Multi-sensor fusion scaffold intelligent monitoring and early warning system and method

Abstract

The invention relates to the technical field of civil engineering safety monitoring, and discloses a multi-sensor fusion scaffold intelligent monitoring and early warning system and method, and the method comprises the steps: deploying a multi-source sensor group at a scaffold key node, collecting data in real time, carrying out the noise reduction fusion processing through an edge calculation module, and constructing a dynamic deformation feature vector; inputting a time sequence prediction model; a self-feedback adjusting unit is triggered to drive an executing mechanism to conduct deformation compensation, and data optimization control parameters are fed back in real time; and based on the compensated deformation state, the remote platform generates a hierarchical maintenance decision and synchronously pushes the hierarchical maintenance decision to the terminal. According to the invention, all-dimensional sensing of deformation, load and environmental parameters of the bridge supporting scaffold is realized through heterogeneous sensor cooperative networking and redundancy check. Local monitoring blind areas can be eliminated, cross validation of a physical model and a data driving algorithm is combined, the reliability of deformation monitoring and the robustness under environment interference are remarkably improved, and more comprehensive data support is provided for construction safety.

Description

Technical Field

[0001] The present invention relates to the technical field of civil engineering safety monitoring, and in particular to a multi-sensor fusion scaffolding intelligent monitoring and early warning system and method. Background Art

[0002] During highway bridge construction, supporting scaffolding serves as a temporary load-bearing structure, and its stability directly impacts construction safety and project progress. With the rapid development of long-span bridge construction, the dynamic and static loads borne by scaffolding structures are becoming increasingly complex. Traditional monitoring methods, which primarily rely on manual inspections and single-sensor fixed-point measurements, suffer from limitations in monitoring dimensions, high data dispersion, and delayed response.

[0003] In the existing technology, the deformation warning method based on static thresholds is difficult to adapt to the dynamic changes of the construction environment, and cannot effectively integrate the impact of multiple factors such as environmental temperature and humidity, mechanical vibration, etc. on structural safety, resulting in a high false alarm rate; at the same time, there is a lack of a quantitative assessment model for the deformation trend of scaffolding throughout its life cycle. Risk judgment mostly relies on experience judgment, making it difficult to accurately identify early hidden deformation.

[0004] Furthermore, most monitoring systems are limited to data collection and alarms, failing to form a closed loop of "perception-decision-control." This results in risk management relying on manual intervention, resulting in inefficient response and insufficient adaptability. Existing technologies struggle to proactively suppress scaffolding deformation and coordinate multi-level responses, particularly in complex geological conditions or extreme weather conditions. This hinders overall improvements in bridge construction safety. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-sensor fusion scaffolding intelligent monitoring and early warning system and method, which solves the problems of single monitoring dimension, delayed risk assessment, passive and extensive control, and insufficient system adaptability in the existing safety monitoring of supporting scaffolding in bridge construction.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: The multi-sensor fusion scaffolding intelligent monitoring and early warning method includes: A multi-source sensor group deployed at the top of the scaffolding poles, foundation grid nodes, and crossbar-pole connection nodes collects vertical displacement, foundation pressure distribution, dual-axis tilt angle, and ambient temperature and humidity data in real time; In the edge computing module, the data is subjected to noise reduction, missing value filling and weighted fusion processing to generate a dynamic deformation feature vector. The dynamic deformation feature vector is input into the time series prediction model, and the warning threshold is dynamically adjusted in combination with environmental parameters to output the scaffolding risk level and warning instructions. The self-feedback adjustment unit is triggered according to the early warning instruction, and the actuator is driven to perform deformation compensation through the composite control algorithm, and the compensation data is fed back to the prediction model; The compensated data is analyzed through the remote monitoring platform, and hierarchical maintenance decision instructions are generated and pushed to the terminal.

[0007] Preferably, the multi-source sensor group includes: The laser rangefinder at the top of the pole transmits vertical displacement data via the RS-485 bus; Distributed pressure sensors embedded in ground-based grid nodes transmit pressure distribution via the LoRa protocol; The dual-axis tilt sensor at the connection point between the crossbar and the vertical pole transmits the tilt angle via the CAN bus; The temperature and humidity sensors on the outside of the scaffolding transmit environmental parameters via the ZigBee protocol.

[0008] Preferably, the noise reduction process includes: Perform wavelet threshold decomposition on the original signal and set the adaptive threshold according to the noise energy distribution; Interpolation is used to fill in missing data segments, and the interpolation interval does not exceed the set multiples of the sampling period.

[0009] Preferably, the weighted fusion process includes: Construct a deformation feature vector including the vertical displacement change rate, foundation pressure gradient and tilt angle change rate; and assign fusion weights based on the accuracy of sensor historical data.

[0010] Preferably, the time series prediction model is an LSTM neural network, including: The sequence of dynamic deformation feature vectors within the input time window; Output the risk probability distribution for future periods; Dynamically adjust the warning threshold based on environmental parameters.

[0011] Preferably, the dynamic adjustment of the warning threshold includes: Calculate the basic threshold based on the mean and standard deviation of historical data; The threshold is linearly corrected according to the changes in ambient temperature and humidity.

[0012] Preferably, the composite control algorithm is fuzzy-PID control, including: Dynamically adjust proportional, integral, and differential parameters according to deformation deviation and change rate; Drive the hydraulic actuator to perform displacement compensation.

[0013] Preferably, the hydraulic actuator: Compensation accuracy is ±0.1mm; The compensation data is fed back to the prediction model through closed-loop feedback.

[0014] Preferably, the hierarchical maintenance decision instruction includes: Level 1 response: Push inspection notification to mobile terminal; Level 2 response: restrict non-operating personnel from entering high-risk areas; Level 3 response: Activate emergency support devices and generate reinforcement plans.

[0015] Multi-sensor fusion scaffolding intelligent monitoring and early warning system, including: Multi-source sensor group: laser rangefinder at the top of the pole, pressure sensor at the foundation grid node, tilt sensor at the connection node, and temperature and humidity sensor on the outside; Edge computing module: Receives laser ranging data via RS-485 bus, pressure data via LoRa, tilt data via CAN bus, and environmental data via ZigBee; performs noise reduction, data fusion, and 3D modeling; Dynamic threshold warning module: receives the deformation feature vector from the edge computing module through the HTTP protocol; outputs warning instructions to the self-feedback adjustment unit; Self-feedback adjustment unit: Receives warning instructions through the MQTT protocol and drives the hydraulic actuator to perform compensation; transmits compensation data back to the edge computing module through the Modbus protocol; Remote monitoring platform: Receives compensated data via TCP / IP protocol; generates maintenance decision instructions and pushes them to terminal devices.

[0016] In summary, the present invention includes at least one of the following beneficial technical effects: 1. This invention achieves full-dimensional perception of bridge support scaffolding deformation, load, and environmental parameters through a collaborative networking of heterogeneous sensors and a redundant verification mechanism. Compared to traditional single-sensor monitoring solutions, this invention's multi-source data fusion technology eliminates local monitoring blind spots. Combining physical models with cross-validation of data-driven algorithms, it significantly improves the reliability of deformation monitoring and its robustness under environmental interference, providing more comprehensive data support for construction safety.

[0017] 2. This invention, based on a fusion assessment method of finite element simulation and an LSTM-GRU hybrid model, overcomes the limitations of traditional static threshold warning systems. By driving real-time iterative calculations of the finite element model and incorporating a dynamic correction mechanism for environmental parameters, it can accurately quantify the risk of structural instability and effectively identify early hidden deformations, achieving a significant improvement from "passive warning" to "active protection." BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the method flow of the present invention; Figure 2 Schematic diagram of the system flow of the present invention. DETAILED DESCRIPTION

[0019] The following is combined with Figure 1 -Attached Figure 2 , the present invention is described in further detail.

[0020] The present invention provides a multi-sensor fusion scaffolding intelligent monitoring and early warning system and method, which may include the following steps: S1. Multi-source sensor data acquisition In this embodiment, the multi-source sensor data acquisition process is based on the structural mechanical properties and construction environment characteristics of the highway bridge supporting scaffolding. Through the heterogeneous sensor collaborative perception network, high-precision synchronous monitoring of the scaffolding deformation, load and environmental parameters throughout the entire bridge construction cycle is achieved.

[0021] In this embodiment, the laser rangefinder's installation position must meet the requirements for monitoring the axial deformation of the bridge support scaffolding uprights. Preferably, the laser rangefinder is secured to the center of the upright's top flange via an adjustable universal bracket, aligning the measurement beam's direction with the upright's axis to eliminate measurement errors introduced by skewed installation. After installation, the laser rangefinder is initially calibrated using a total station to ensure its reference coordinate system aligns with the bridge construction coordinate system.

[0022] In the scaffolding foundation's cap area, the gridded deployment of the pressure sensor array must be integrated with the load transfer path of the bridge piers. Optimally, the grid spacing is dynamically adjusted based on the stress diffusion angle determined by finite element simulation: a 0.3m x 0.3m dense grid is used within a 0.5m radius around the piers, gradually transitioning to a 1.0m x 1.0m standard grid outward. The sensitive surface of each pressure sensor is flush with the foundation concrete surface and encapsulated with epoxy resin to prevent mechanical damage during concrete pouring.

[0023] In this embodiment, a three-axis MEMS inclination sensor is installed at the connection point between the crossbar and the pier to prevent the horizontal offset and torsional deformation of the bridge support scaffolding. The sensor's X-axis and Y-axis are aligned with the longitudinal and transverse directions of the bridge, respectively, and the Z-axis is perpendicular to the bridge deck. During installation, a dual fixing method of a magnetic base and mechanical locking is used to suppress sensor displacement caused by construction vibration. Preferably, the inclination sensor has a built-in temperature compensation unit, which corrects the temperature drift error using the following formula: θ 校正 =θ 原始 -k T ·(TT 标定 ); Among them, k T is the temperature sensitivity coefficient, T 标定 This is the factory calibration temperature, and compensation is automatically performed by the sensor firmware.

[0024] In this embodiment, the deployment position of the vibration acceleration sensor needs to reflect the transfer characteristics of the dynamic load of the bridge construction. Preferably, an ICP type acceleration sensor is installed at the mid-span position of the scaffolding beam below the concrete pouring area of ​​the bridge deck, with the sensitive direction perpendicular to the bridge deck to capture the vertical vibration component. In order to suppress high-frequency noise interference, the sensor has a built-in mechanical low-pass filter, and the cut-off frequency is set to 2 times the main frequency of the construction machinery excitation. In this embodiment, the installation height of the ultrasonic anemometer is consistent with the elevation of the top surface of the bridge main beam to avoid wind speed measurement distortion caused by the flow effect around the bridge tower. Preferably, the wind vane of the anemometer adopts a bearingless magnetic levitation design to ensure that 360° continuous angle measurement can still be achieved in a strong wind environment. The probe of the temperature and humidity sensor is covered with a radiation shield, and the air flow rate around the probe is maintained at ≥1m / s by a forced convection fan to eliminate the temperature measurement deviation caused by solar radiation.

[0025] In this embodiment, a multi-mode fusion transmission strategy is adopted to meet the communication requirements of large-span bridge construction scenarios: The laser rangefinder and the inclination sensor are networked via the RS-485 bus, using the Modbus-RTU protocol to achieve 10ms-level synchronous acquisition in a master-slave polling mode. The pressure sensor array is self-organized through the LoRa wireless network and uses the TDMA time slot allocation algorithm to ensure that the data collision rate under large-scale nodes is less than 0.1%; The vibration acceleration sensor uses a coaxial cable to transmit analog signals, and a 24-bit Σ-Δ ADC is integrated in the edge computing module for digital processing.

[0026] In this embodiment, dual-channel redundant transmission of key deformation parameters is achieved through the following steps: The data on the primary channel (wired) and the backup channel (4G wireless) are time-aligned, and interpolation is used to compensate for transmission delay differences. The dynamic time warping (DTW) distance between the two channel data is calculated. If the DTW distance exceeds a preset threshold, the primary channel is deemed faulty. The fault isolation mechanism is triggered, the backup channel is upgraded to the primary channel, and the abnormal data segment is marked for subsequent analysis.

[0027] In this embodiment, the clock synchronization of distributed sensors is ensured by the following process: 1. The edge computing module periodically broadcasts IEEE 1588 Precision Time Protocol (PTP) synchronization messages; 2. Each sensor calculates the clock offset based on the message transmission delay and corrects the local clock; 3. For nodes that cannot directly receive PTP messages, a clock drift prediction model based on Bayesian inference is used, and the formula is: Δt i =α·Δt i-1 +β·(T晶振 -T 环境 ); Where α is the historical offset weight coefficient, and β is the temperature drift compensation coefficient, which are determined by offline calibration.

[0028] In this embodiment, the electromagnetic compatibility design of the sensor node includes: The pressure sensor signal conditioning circuit uses a differential input structure and shielded twisted pair to suppress common-mode interference; The vibration acceleration sensor power supply circuit integrates a π-type filter to attenuate the high-frequency noise of the switching power supply; The optical window of the laser rangefinder is coated with ITO conductive film to prevent measurement distortion caused by static electricity accumulation.

[0029] Adaptive deployment strategy In this embodiment, the sensor layout density optimization method includes: 1. Calculate the deformation sensitivity coefficient of each node based on the finite element model of the bridge support scaffolding (U is the structural strain energy, x i is the node coordinate); 2. For areas where the sensitivity coefficient is higher than the threshold, the deployment density is increased to three times the baseline value; 3. Dynamic adjustment in combination with the construction stage: focus on monitoring the mid-span area during the concrete pouring period, and focus on monitoring the anchor points during the tensioning period.

[0030] In this embodiment, the sensor layout density optimization method includes: 1. Calculate the deformation sensitivity coefficient of each node based on the finite element model of the bridge support scaffolding (U is the structural strain energy, x i is the node coordinate); 2. For areas where the sensitivity coefficient is higher than the threshold, the deployment density is increased to three times the baseline value; 3. Dynamic adjustment in combination with the construction stage: focus on monitoring the mid-span area during the concrete pouring period, and focus on monitoring the anchor points during the tensioning period.

[0031] Data integrity assurance In this embodiment, the data integrity verification process based on the graph neural network (GNN) is as follows: 1. Construct the scaffolding node topology graph G = (V, E), where the vertex V represents the sensor node and the edge E represents the mechanical connection relationship; 2. Define the message passing function m ij =f(h i ,h j ,e ij ), aggregate adjacent node features h i ,h j With edge attribute e ij ; 3. Predict the probability of abnormal nodes through multi-layer graph convolution operations. If the probability exceeds the threshold, start the data repair program.

[0032] In this embodiment, the multi-source sensor data acquisition solution completely covers the monitoring needs in bridge construction scenarios through the collaborative design of spatial deployment optimization, heterogeneous communication integration, redundant verification mechanism and intelligent fault-tolerant algorithm, providing high-reliability and high-precision raw data input for subsequent data processing and safety warning.

[0033] S2, edge computing data processing In this embodiment, the edge computing data processing process performs multi-level processing and feature fusion on the raw data collected by multi-source sensors, and realizes noise suppression, data repair and multi-dimensional feature extraction through adaptive algorithms, providing high signal-to-noise ratio input features for subsequent risk assessment models.

[0034] Data denoising and signal reconstruction In this embodiment, a noise suppression method based on wavelet packet decomposition is used to address the complex electromagnetic interference and mechanical vibration noise in bridge construction scenarios. Specifically, the original signal is subjected to multi-layer wavelet packet decomposition, and the threshold of each subband is dynamically set according to the noise energy distribution characteristics. Preferably, the threshold is calculated as follows: Among them, σ j is the noise standard deviation of the j-th layer wavelet coefficient, calculated by the robust estimation method (Median Absolute Deviation, MAD); N j is the coefficient length of this layer. For the non-stationary noise in the bridge vibration signal, the translation-invariant wavelet transform (TIWFT) is further introduced to suppress the pseudo-Gibbs phenomenon through cyclic translation operation and improve the smoothness of signal reconstruction.

[0035] Missing data repair and interpolation optimization In this embodiment, a segmented interpolation strategy is used to address local data loss caused by sensor occlusion or communication interruption. Preferably, for short-term missing segments (missing duration ≤ 2 times the sampling period), cubic spline interpolation is used to fill in the gaps and ensure the continuity of the second-order derivative of the interpolation curve. For long-term missing segments, virtual data is generated through finite element simulation based on the mechanical properties of the scaffolding structure as interpolation constraints. The formula is: Ku=f; Among them, K is the scaffolding stiffness matrix, u is the node displacement vector, and f is the load vector. The interpolation sequence under physical constraints is generated through iterative calculation.

[0036] Multi-source data fusion and weight distribution In this embodiment, the feature fusion process is based on sensor confidence and dynamic correlation analysis. Preferably, the deformation feature vector is constructed in: ΔL is the rate of change of vertical displacement, calculated by the difference method; is the foundation pressure gradient, which is solved using the central difference operator based on the gridded pressure data; is the rate of change of the tilt angle, obtained by linear fitting within the sliding time window. The fusion weight distribution adopts a hybrid weight model, which comprehensively considers the historical accuracy and real-time status of the sensor: in, is the historical variance of the sensor, reflecting the long-term stability; corr(f i ,D ref ) is the correlation coefficient between the current feature vector and the reference deformation database, which is used to capture dynamic correlation; β is the adjustment factor, which is used to balance the historical accuracy and real-time state weight. Time series feature extraction and model adaptation.

[0037] In this embodiment, the built-in time series prediction model of the edge computing module adopts an LSTM network architecture, and its input layer receives a dynamic deformation feature sequence within a time window. Preferably, the length of the time window is dynamically adjusted according to the bridge construction stage: a short window (such as 10 seconds) is used during the concrete pouring period to capture rapid deformation, and a long window (such as 60 seconds) is used during the maintenance period to extract low-frequency trend features. The network hidden layer is designed as a bidirectional LSTM structure, which captures the time series dependency through forward and backward propagation, and the output layer generates the risk level probability distribution through the Softmax function.

[0038] Dynamic threshold adjustment and parameter coupling: In this embodiment, the dynamic adjustment strategy of the warning threshold is based on the coupling mechanism of environmental parameters and structural response. Specifically, the temperature and humidity changes are introduced into the threshold correction through the following formula: τ(t)=μ(t)+2σ(t)+α1·ΔT+α2·ΔRH; Among them, μ(t) and σ(t) are the mean and standard deviation within the sliding time window, reflecting the baseline characteristics of structural deformation; ΔT and ΔRH are the changes in temperature and humidity, which are input after being filtered by the exponential smoothing method; α1 and α2 are coupling coefficients, which are calibrated according to the thermal expansion coefficient and humidity sensitivity of the bridge material.

[0039] Model online update and incremental learning: In this embodiment, the edge computing module supports online updating of model parameters. Preferably, when the compensated deformation data is transmitted back to the edge node, the LSTM network weights are updated using an incremental learning algorithm: Freeze the weights of the feedforward part of the network and only fine-tune the output layer and the last two LSTM layers; The elastic weight consolidation (EWC) algorithm is used to constrain the weight update amplitude to prevent catastrophic forgetting. The updated model parameters are synchronized to the cloud for version management and consistency verification.

[0040] In this embodiment, the task scheduling of the edge computing module adopts a dynamic priority allocation mechanism. Preferably, the data processing tasks are divided into real-time tasks (such as noise reduction and interpolation) and non-real-time tasks (such as model training), and the deadline constraints of real-time tasks are ensured by preemptive scheduling. For high-concurrency scenarios, lightweight model distillation technology is used to compress the full LSTM network into a micro GRU network to maintain basic early warning functions when computing resources are limited.

[0041] In this embodiment, the edge computing data processing solution realizes the full-process processing from raw data to risk assessment features through the collaborative design of noise suppression, data repair, multi-source fusion and model optimization, providing high-reliability and low-latency computing support for bridge construction safety monitoring.

[0042] S3. Dynamic risk assessment and early warning In this embodiment, the dynamic risk assessment and early warning process is based on multi-source sensor data and deformation features extracted by edge computing. By integrating physical models and machine learning algorithms, real-time quantitative assessment and graded early warning of bridge support scaffolding structural risks are achieved, providing data-driven support for construction safety decisions.

[0043] In this embodiment, the risk assessment model uses a hybrid architecture of finite element simulation and deep learning. Preferably, a parametric finite element model is pre-installed in the edge computing module. The model automatically generates geometric parameters such as vertical bar spacing, horizontal bar pitch, and bracing angle based on the scaffolding design drawings, and imports material constitutive relations and boundary conditions. By inputting the vertical displacement, foundation pressure, and inclination data collected by the sensor in real time, the finite element model is driven to iteratively calculate the structural strain energy distribution and critical instability load, generating a deformation prediction benchmark under physical constraints.

[0044] The coupling between the finite element model and the data-driven model is achieved through a residual learning mechanism. Specifically, the residual between the output of the LSTM-GRU hybrid network and the finite element simulation results is used as the final risk assessment indicator: P failure =f AI (F seq )+λ·(U FEM -U threshold ); Among them, f AI is the instability probability of the hybrid neural network output, U FEMis the strain energy density calculated by finite element method, λ is the weight coefficient of the physical model, and its value range is determined through offline training.

[0045] Time series prediction model architecture optimization: In this embodiment, the LSTM-GRU hybrid network is structurally optimized for the temporal characteristics of the bridge construction scene. The input layer receives the deformation feature vector sequence F in the time window. seq =[F t-T ,…,F t ], which includes dynamic features such as the rate of change of vertical displacement, foundation pressure gradient, and rate of change of inclination angle. Preferably, the length of the time window is dynamically adjusted according to the construction stage: a short window (e.g., 30 seconds) is used during the concrete pouring period to capture rapid deformation, and a long window (e.g., 300 seconds) is used during the curing period to extract low-frequency trends.

[0046] The network's hidden layer is designed as a cascade of bidirectional LSTM and GRU: the forward LSTM layer captures historical temporal dependencies, the reverse GRU layer extracts future potential correlations, and multi-scale features are integrated through skip connections. The output layer uses an attention mechanism to weight the contribution of each time step. The formula is: α t =softmax(v T tanh(W h h t +W c c t )); Among them, h t is the LSTM hidden state, c t is the GRU cell state, W h ,W c is a trainable weight matrix, v is the attention vector, and the weighted features are mapped to the risk probability space through a fully connected layer.

[0047] Dynamic threshold correction and environmental coupling: In this embodiment, the dynamic adjustment strategy of the warning threshold comprehensively considers the coupling effect of structural response and environmental parameters. Basic threshold τ base The critical strain energy density output from the finite element model is converted to: Among them, k1 and k2 are the material property correlation coefficients, which are determined by laboratory calibration. The environmental parameter correction term introduces the real-time changes in temperature, humidity and wind speed: in, is the material thermal expansion coefficient and humidity sensitivity, θ is the angle between the wind direction and the bridge axis, and α1, α2, and α3 are the environmental coupling coefficients. The final correction threshold τ final =τ base +Δτ.

[0048] Multi-objective optimization and risk quantification In this embodiment, the risk level classification is based on a multi-objective optimization framework, taking into account the probability of instability, deformation rate and environmental interference intensity. Define the objective function: in, is a comprehensive indicator of environmental interference. The weight coefficients w1, w2, and w3 are determined by the analytic hierarchy process (AHP). The risk is divided into three levels according to the objective function value: Level 1 (low risk): The objective function value is lower than the 10% quantile of historical data; Level 2 (medium risk): the objective function value is between the 10% and 90% quantiles; Level 3 (high risk): The objective function value exceeds the 90th percentile.

[0049] Tiered Response and Decision Making: In this embodiment, after the early warning instruction is generated, a hierarchical response is achieved through the following mechanism: 1. Level 1 response: Triggering local data resampling in the edge computing module, increasing the frequency of sensor data collection in risk areas, and simultaneously pushing inspection work orders to mobile terminals. 2. Secondary response: Activate the preload adjustment mode of the hydraulic actuators, limit the movement speed of bridge deck construction equipment, and identify high-risk areas using the BIM model. 3. Level 3 response: Interrupt concrete pouring operations, activate the hydraulic locking device of the emergency support truss, and synchronize the traffic control system to evacuate vehicles under the bridge.

[0050] Model online learning and robustness enhancement: In this embodiment, the risk assessment model supports incremental learning to adapt to the dynamic changes in bridge construction. When the hydraulic actuator feedback compensation data, the model is updated through the following steps: 1. Construct the data comparison set before and after compensation D = {(F before ,F after )}; 2. Calculate the feature offset ΔF = F after -F before ; 3. Use Generative Adversarial Networks (GANs) to synthesize extended training samples and enhance the model’s generalization ability to compensated states; 4. Based on the elastic weight consolidation (EWC) algorithm, the network weight update is constrained to retain historical knowledge to prevent forgetting.

[0051] In this embodiment, the dynamic risk assessment and early warning solution uses core technologies such as physical-data model fusion, environmental coupling threshold correction, and multi-objective optimization classification to achieve accurate quantification and active prevention and control of bridge support scaffolding risks, providing full-cycle protection for structural safety in complex construction scenarios.

[0052] S4, closed loop variable suppression and feedback In this embodiment, the closed-loop deformation suppression and feedback process is based on the early warning instructions output by the dynamic risk assessment module. The hydraulic actuator is driven by a multi-objective optimization control algorithm to achieve active deformation compensation, and the compensation effect is fed back to the prediction model in real time, forming a closed-loop control link of "perception-decision-execution-verification" to ensure the structural stability of the bridge support scaffolding.

[0053] Fuzzy-PID compound control algorithm: In this embodiment, the closed-loop control algorithm adopts a composite architecture combining fuzzy logic and PID control to cope with the nonlinear and time-varying characteristics of the deformation process during bridge construction. Specifically, the fuzzy reasoning module calculates the deformation deviation e(t) and its rate of change based on the real-time deformation deviation e(t). Dynamically adjust the proportional coefficient K of the PID controller p , integration time T i and differential time T d Preferably, the fuzzy rule base is constructed based on the experimental data of scaffold deformation response, and defines the input variable e(t) and The membership function is Gaussian distribution, and the output variable ΔK p , ΔT i , ΔT d The membership function is triangular distribution, and the center of gravity method is used to solve the fuzzy problem and generate precise control parameters: K p (t) = K p0 +ΔK p ,T i (t) = T i0 +ΔT i ,T d (t) = T d0 +ΔT d ; Among them, K p0 、T i0 、T d0 are the initial PID parameters, which are determined by the Ziegler-Nichols tuning method.

[0054] Dynamic compensation of hydraulic actuators: In this embodiment, the hydraulic actuator drives the displacement adjustment of the target pole according to the compensation amount u(t) output by the control algorithm. Preferably, the hydraulic cylinder adopts a ball screw transmission mechanism driven by a servo motor, and its displacement output model is: Among them, A p is the effective area of ​​the piston, F fric is the friction resistance, F load is the reverse load applied by the scaffolding, and m is the equivalent moving mass. To suppress overshoot, the hydraulic system features built-in pressure-flow dual closed-loop control, with proportional valves regulating oil flow to ensure displacement tracking accuracy better than ±0.1 mm.

[0055] Multi-node collaborative compensation strategy In this embodiment, a distributed collaborative control strategy is adopted for the multi-point coupling deformation characteristics of the bridge support scaffolding. Preferably, the compensation amount x of the target pole is defined as target The coordinated compensation amount x with its adjacent pole adj Satisfy the mechanical equilibrium constraints: Among them, k i is the equivalent stiffness coefficient of the vertical pole, x nominal The optimal compensation combination is solved by the sequential quadratic programming (SQP) algorithm to minimize the overall deformation energy. Energy consumption of hydraulic system The weighted sum of the construction phase is dynamically adjusted according to the construction stage.

[0056] Compensation effect verification and model update: In this embodiment, the structural state after deformation compensation is fed back to the edge computing module in real time through the multi-source sensor network. Preferably, a Kalman filter is used to fuse the displacement sensor data and the tilt sensor data to estimate the actual deformation amount ΔH after compensation. actual and the target compensation amount ΔH ref Compare and generate the residual signal: ∈=ΔH ref -ΔH actual ; When the residual exceeds the preset threshold, the control parameter self-tuning program is triggered: an adaptive law is designed based on Lyapunov stability theory, and the proportional coefficient and integral time of the fuzzy-PID controller are dynamically adjusted to ensure the stability of the closed-loop system.

[0057] Data closure and knowledge accumulation: In this embodiment, the operational data generated during the compensation process (such as hydraulic pressure, valve core displacement, and actual compensation amount) is stored in the edge database, and the system performance is optimized through the following mechanisms: Iterative optimization of control parameters: Use reinforcement learning algorithms (such as DDPG) to train control strategies offline and generate parameter configuration tables that adapt to different construction stages; Finite element model modification: Inversely calculate the scaffolding equivalent stiffness parameters based on the compensated deformation data and update the material constitutive relationship in the finite element model; Fault knowledge base construction: Record the handling strategies for abnormal events such as actuator overruns and communication delays to improve the system's fault tolerance.

[0058] In this embodiment, the closed-loop deformation suppression and feedback scheme achieves precise suppression and continuous optimization of the deformation of the bridge support scaffolding through the collaborative design of a composite control algorithm, a collaborative drive strategy, and an adaptive learning mechanism, providing active protection for structural safety in complex construction environments. S5. Graded Response and Decision Making In this embodiment, the hierarchical response and decision-making process is based on the dynamic risk assessment results. Differentiated maintenance instructions are generated through multi-source data fusion and intelligent decision-making algorithms, and closed-loop control is achieved by linking the actuators and management systems to ensure full-cycle risk management of bridge support scaffolding and optimal decision-making for construction safety.

[0059] Risk level classification and response rules In this embodiment, the risk level classification is based on multi-dimensional indicators including instability probability, deformation trend and environmental interference intensity. Preferably, a fuzzy comprehensive evaluation model is used to calculate the comprehensive risk index: Among them, P failure is the probability of instability, is the gradient norm of the deformation feature vector, I env It is a comprehensive indicator of environmental interference. The weight coefficients w1, w2, and w3 are calibrated by expert scoring and historical data. The response level is divided into three levels according to the risk index threshold: Level 1 response: triggers local monitoring enhancement and manual inspection instructions; Level 2 response: Initiate active deformation suppression and construction restriction measures; Level 3 response: Implement emergency reinforcement and traffic control linkage.

[0060] Multi-source data fusion decision-making: In this embodiment, decision instructions are generated based on a collaborative analysis of sensor data, finite element simulation results, and a historical case library. Preferably, a decision-making knowledge graph is constructed, where nodes include entities such as risk type, treatment measure, and construction phase, and edge attributes describe the association rules between entities. When the risk index exceeds the limit, a graph neural network (GNN) traverses the knowledge graph to generate a set of candidate solutions that include treatment measure priorities, resource requirements, and expected effects.

[0061] Dynamic Priority Scheduling and Resource Allocation In this embodiment, the execution priority of the response instruction is dynamically adjusted according to the risk propagation speed and construction progress. Preferably, the risk propagation coefficient is defined as: in, is the risk index change rate, is the spatial gradient of strain energy. For areas with high propagation coefficients, priority is given to scheduling adjacent hydraulic actuators and inspection resources. Resource allocation uses an integer programming model, and the objective function is to minimize response delay and resource consumption: Among them, x i Enable flag for actuator, y i is the human resource allocation, c i d i To correspond to the cost coefficient, the constraints include construction specification requirements and equipment availability.

[0062] Multi-level response coordinated execution: In this embodiment, the first-level response instructions include increasing the sampling frequency of local sensors, generating an electronic inspection work order, and pushing it to the manager's mobile terminal. Preferably, the electronic work order automatically populates the risk location, inspection items, and operating procedures using natural language generation (NLG) technology, and is linked to BIM model coordinates to support AR navigation and positioning.

[0063] The secondary response command triggers active deformation suppression measures: the compensation amount of the hydraulic actuator is adjusted through the fuzzy-PID controller, while limiting the movement speed and operating range of the bridge deck construction equipment. Preferably, the construction restriction strategy is designed based on the game theory model to balance safety constraints and construction efficiency: Among them, S safety is the security utility function, S efficiency is the construction efficiency function, and α is the risk adaptive coefficient, which is dynamically adjusted with the risk index.

[0064] The third-level response command activates the emergency support device and links to the traffic control system. The deployment logic of the emergency support truss is preferably based on the distribution of deformation energy density: hydraulic jacking devices are prioritized in areas with high strain energy, and the optimal support angle and preload force are calculated through finite element model inversion. Traffic control commands are transmitted to the roadside unit via the V2X communication protocol, dynamically updating the electronic fence range and vehicle diversion path.

[0065] In this embodiment, maintenance decision instructions are generated by integrating historical maintenance records with real-time operating data. Preferably, a sequence-to-sequence (Seq2Seq) model is used to map risk signatures to standardized work order templates, outputting structured instructions that include construction personnel, equipment requirements, operational procedures, and acceptance criteria. After the work order is executed, structural status data is collected through a sensor network to verify the effectiveness of the measures and generate a closed-loop feedback report.

[0066] Self-optimization of the decision model: In this embodiment, the maintenance decision knowledge base supports online updates and incremental learning. When new work order execution data is fed back to the system, the decision model is optimized through the following steps: Extract work order execution effect characteristics (such as deformation suppression rate and downtime reduction); Calculate the similarity between historical decisions and current decisions, and use contrastive learning algorithms to mine potential association rules; Update the knowledge graph edge weights and decision model parameters to improve the adaptation accuracy of subsequent instructions.

[0067] In this embodiment, the hierarchical response and decision-making scheme realizes the closed-loop optimization of full-factor control of bridge support scaffolding risks and construction safety decision-making through the collaborative design of multi-dimensional risk assessment, dynamic resource scheduling and intelligent instruction generation, providing a systematic solution for emergency response in complex engineering scenarios.

[0068] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The multi-sensor fusion scaffolding intelligent monitoring and early warning method is characterized by: include: A multi-source sensor group deployed at the top of the scaffolding poles, foundation grid nodes, and crossbar-pole connection nodes collects vertical displacement, foundation pressure distribution, dual-axis tilt angle, and ambient temperature and humidity data in real time; In the edge computing module, the data is subjected to noise reduction, missing value completion, and weighted fusion processing to generate dynamic deformation feature vectors; The dynamic deformation feature vector is input into the time series prediction model, and the warning threshold is dynamically adjusted in combination with environmental parameters to output the scaffolding risk level and warning instructions; The self-feedback adjustment unit is triggered according to the early warning instruction, and the actuator is driven to perform deformation compensation through the composite control algorithm, and the compensation data is fed back to the prediction model; The compensated data is analyzed through the remote monitoring platform, and hierarchical maintenance decision instructions are generated and pushed to the terminal.

2. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: The multi-source sensor set includes: The laser rangefinder at the top of the pole transmits vertical displacement data via the RS-485 bus; Distributed pressure sensors embedded in ground-based grid nodes transmit pressure distribution via the LoRa protocol; The dual-axis tilt sensor at the connection point between the crossbar and the vertical pole transmits the tilt angle via the CAN bus; The temperature and humidity sensors on the outside of the scaffolding transmit environmental parameters via the ZigBee protocol.

3. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: Noise reduction processing includes: Perform wavelet threshold decomposition on the original signal and set the adaptive threshold according to the noise energy distribution; Interpolation is used to fill in missing data segments, and the interpolation interval does not exceed the set multiples of the sampling period.

4. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: Weighted fusion processing includes: Construct a deformation feature vector including the vertical displacement change rate, foundation pressure gradient and tilt angle change rate; Fusion weights are assigned based on the accuracy of sensor historical data.

5. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: The time series prediction model is an LSTM neural network, including: The sequence of dynamic deformation feature vectors within the input time window; Output the risk probability distribution for future periods; Dynamically adjust the warning threshold based on environmental parameters.

6. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: Dynamic adjustment of warning thresholds includes: Calculate the basic threshold based on the mean and standard deviation of historical data; The threshold is linearly corrected according to the changes in ambient temperature and humidity.

7. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: The composite control algorithm is fuzzy-PID control, including: Dynamically adjust proportional, integral, and differential parameters according to deformation deviation and change rate; Drive the hydraulic actuator to perform displacement compensation.

8. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 7 is characterized in that: Hydraulic actuator: Compensation accuracy is ±0.1mm; The compensation data is fed back to the prediction model through closed-loop feedback.

9. The multi-sensor fusion scaffolding intelligent monitoring and early warning method according to claim 1 is characterized in that: Hierarchical maintenance decision instructions include: Level 1 response: Push inspection notification to mobile terminal; Level 2 response: restrict non-operating personnel from entering high-risk areas; Level 3 response: Activate emergency support devices and generate reinforcement plans.

10. A multi-sensor fusion scaffolding intelligent monitoring and early warning system, according to the multi-sensor fusion scaffolding intelligent monitoring and early warning method according to any one of claims 1 to 9, characterized in that: include: Multi-source sensor group: laser rangefinder at the top of the pole, pressure sensor at the foundation grid node, tilt sensor at the connection node, and temperature and humidity sensor on the outside; Edge computing module: Receives laser ranging data via RS-485 bus, pressure data via LoRa, tilt data via CAN bus, and environmental data via ZigBee; performs noise reduction, data fusion, and 3D modeling; Dynamic threshold warning module: receives the deformation feature vector from the edge computing module through the HTTP protocol; Outputting warning instructions to the self-feedback adjustment unit; Self-feedback adjustment unit: Receives warning instructions through the MQTT protocol and drives the hydraulic actuator to perform compensation; transmits compensation data back to the edge computing module through the Modbus protocol; Remote monitoring platform: Receives compensated data via TCP / IP protocol; generates maintenance decision instructions and pushes them to terminal devices.

Images