Multi-source sensing based formwork system stress real-time monitoring and early warning system and method thereof

Abstract

The present application relates to the technical field of building formwork support, and particularly relates to a formwork support system stress real-time monitoring and early warning system and method based on multi-source sensing, comprising a multi-source sensor network, a data acquisition and transmission module, and a data processing and calculation module; compared with the prior art which generally adopts a rigid alarm mechanism based on a fixed threshold, the essence of the prior art is a post-response mode and cannot quantify the risk level; the present application introduces a probabilistic finite element analysis method, processes input parameters such as material properties and load conditions as probability distributions, generates a probability distribution interval of key mechanical responses through Monte Carlo simulation, and finally accurately quantifies the risk level by calculating the tail probability of measured data falling outside the probability interval; this early warning mechanism based on probability statistics not only significantly reduces the false alarm rate, but also realizes a substantial leap from pure alarm to risk level assessment.

Description

Technical Field

[0001] This invention relates to the field of building formwork technology, and in particular to a real-time monitoring and early warning system and method for stress on formwork systems based on multi-source sensing. Background Technology

[0002] Formwork systems are critical temporary support structures in concrete construction, and their safety directly affects the overall quality of the project. In recent years, sensor-based real-time monitoring technology has become an important means of ensuring the safety of formwork systems.

[0003] Existing monitoring technologies mainly employ fixed threshold alarm mechanisms, which have significant shortcomings: First, rigid thresholds are difficult to adapt to dynamic changes during construction, easily leading to false alarms and missed alarms; second, they can only provide binary alarms and cannot quantify risk levels; third, they rely on isolated data for judgment and lack the ability to fuse multi-source data and analyze fault mechanisms; fourth, fixed sampling strategies result in energy waste or monitoring blind spots; and fifth, simulation models are disconnected from actual conditions, making it difficult to maintain accuracy.

[0004] This invention constructs an intelligent monitoring and early warning system through multi-source sensing and artificial intelligence technologies. It employs parameter inversion technology for dynamic model calibration, uses probabilistic finite element analysis for quantitative risk assessment, utilizes reinforcement learning to optimize sampling strategies, combines generative AI to reconstruct the entire field's mechanical state, and achieves fault mechanism analysis and proactive early warning through causal discovery and active detection mechanisms, ultimately forming an intelligent monitoring solution. Summary of the Invention

[0005] To overcome the problems mentioned in the background art, the present invention proposes a real-time force monitoring and early warning system and method for a formwork support system based on multi-source sensing.

[0006] The technical solution of this invention is: a real-time force monitoring and early warning system for a formwork support system based on multi-source sensing, comprising:

[0007] A multi-source sensor network deployed at key nodes of the formwork support system;

[0008] Data acquisition and transmission module connected to sensor network;

[0009] The data processing and calculation module, which communicates with the data acquisition and transmission module, specifically includes:

[0010] The parametric finite element simulation module is used to simulate the mechanical behavior of the formwork system.

[0011] The model update module is used to update the simulation model using real-time sensor data.

[0012] An adaptive sampling control module is used to dynamically adjust the sampling strategies of each sensor;

[0013] The data augmentation module is used to generate full-field mechanical state information based on sparse sensor data;

[0014] The causal analysis and knowledge base module is used to identify the causal relationships of faults and store fault modes.

[0015] The probability analysis and early warning module is used to perform probabilistic mechanical analysis and output the risk probability;

[0016] An active detection module is used to propose risk hypotheses and control sensors to verify them.

[0017] Preferably, the parametric finite element simulation module, when simulating the mechanical behavior of the formwork system, specifically includes:

[0018] S11: Model building. Based on the geometric topology, material property parameter set, load condition parameter set, and boundary condition parameter set of the support system, a parameterized finite element model is established. The parameters in the material property parameter set, load condition parameter set, and boundary condition parameter set are dynamically adjustable variables.

[0019] S12: Element Calculation. This step iterates through each element in the finite element model, calculating the element stiffness matrix for each element based on its type, geometric properties, and current material properties, using the finite element theory. The underlying formula is as follows:

[0020] ;

[0021] in, The element stiffness matrix, The strain-displacement matrix, This is a matrix transpose operation. The elastic matrix is ​​used to characterize the constitutive relation of a material. Unit volume;

[0022] S13: Assembly of the overall stiffness matrix. The element stiffness matrices of all elements are assembled into the overall stiffness matrix of the system according to the global number of their respective nodes.

[0023] S14: Load vector assembly, assembling the overall load vector of the system based on the current load condition parameter set;

[0024] S15: Solve the equilibrium equations. Solve the system equilibrium equations consisting of the global stiffness matrix, global displacement vector, and global load vector to obtain the displacement solutions for all nodes. The relationship between strain and displacement at a point within an element is expressed by the strain-displacement matrix as follows:

[0025] ;

[0026] in, For the strain at a point within the unit, This is the displacement vector of all nodes in this element;

[0027] S16: Result Derivation. Based on the nodal displacement vectors obtained from the solution, the stress, strain, and other derived mechanical responses of each element and node are calculated and output according to the element shape function and material constitutive relation. The stress at a certain point inside the element is calculated through the constitutive relation.

[0028] ;

[0029] in, It is the elasticity matrix.

[0030] Preferably, when updating the simulation model using real-time sensor data, the model update module updates the simulation model based on parameter inversion technology, specifically including:

[0031] S21: Define the parameter set and objective function. Select the key parameters to be inverted from the parametric finite element simulation module to form a vector, and define the objective function based on the deviation between the sensor measured data and the simulation data. The expression of the objective function is as follows:

[0032] ;

[0033] in, Let be the objective function. For the measured data of the i-th sensor, This refers to the mechanical response output of the simulation model, corresponding to the sensor's location and type, under the current parameter vector. Let n be the vector of parameters to be inverted, and n be the number of sensors. The importance weight of the i-th sensor data;

[0034] S22: Iterative optimization solution, using optimization algorithms, takes minimizing the objective function as the criterion, and automatically iteratively adjusts the parameter vector within the feasible region of the parameters to find its optimal solution;

[0035] S23: Model update, feeding back the obtained optimal parameter solution to the parametric finite element simulation module, replacing the original parameters, and completing a calibration update of the simulation model.

[0036] As a preferred embodiment, the adaptive sampling control module incorporates a reinforcement learning agent. The state space of the reinforcement learning agent is based on the health status index and force complexity index of the current support system, the action space consists of adjustment instructions for the sampling frequency and accuracy of each sensor, and the reward function is a comprehensive evaluation of risk identification capability and total system energy consumption. The specific workflow of the adaptive sampling control module is as follows:

[0037] S31: State perception, acquires and processes real-time data from multi-source sensor networks and analysis results from parameterized finite element simulation modules, and constructs the state vector of the reinforcement learning agent;

[0038] S32: Policy decision-making. Based on the current state, the agent outputs an action vector according to its policy network. The output action vector is an adjustment instruction for the sampling frequency and sampling accuracy of each sensor in the network in the next time window.

[0039] S33: Action execution, sending the action vector to the data acquisition and communication module to dynamically configure the sampling parameters of each sensor;

[0040] S34: Reward Calculation. After a time period, the reward is calculated based on the new state. The reward is used to evaluate the quality of the action. The principle formula is as follows:

[0041] ;

[0042] in, For instant rewards, For the reward function, This is a risk identification reward function, whose value is positively correlated with the degree of abnormal risk discovered in the new state. This is an energy consumption penalty function, the value of which is positively correlated with the increase in the total power consumption of the system after the action is performed. To reinforce the learning agent's state vector, To reinforce the learning agent's action vectors, For risk identification reward weighting coefficient, This is the energy consumption penalty weighting coefficient;

[0043] S35: Policy update, store the state transition sequence in the experience replay buffer, and periodically sample data to train and update the policy network to optimize long-term cumulative rewards.

[0044] Preferably, when generating full-field mechanical state information based on sparse sensing data, the data augmentation module specifically includes:

[0045] S41: Data preparation and preprocessing: Obtain sparse sensor measured data and corresponding full-field mechanical field data output by the parameterized finite element simulation module, normalize the data, and construct a training dataset.

[0046] S42: Model training, using sensor data as a condition and the corresponding simulated full-field data as the target, trains a generative AI model with the goal of learning the mapping relationship from sensor data to simulated full-field data.

[0047] S43: High-resolution data generation. When the system is running online, the real-time collected sparse sensor data is input into the pre-trained generative AI model to generate high-resolution full-field mechanical state data. The values ​​of the full-field mechanical state data at the sensor points are consistent with the real-time collected sparse sensor data, and physically reasonable interpolation and extrapolation values ​​are generated in other areas.

[0048] As a preferred approach, generative AI models are implemented using robust generative adversarial networks, including:

[0049] The generator takes random noise vectors and sparse sensor measured data as inputs to generate high-resolution full-field mechanical state data.

[0050] The discriminator takes the generated data output by the generator and the simulation data output by the parameterized finite element simulation module as inputs, and uses the measured data from the sparse sensor as a condition to determine the authenticity of the input data.

[0051] The generator and discriminator are trained adversarially using the following objective function:

[0052] ;

[0053] in, Let E be the objective function, where E represents the expected value, x is the actual full-field data sample, and y is the conditional data. For discriminator, Let z be the generator, and z be a random noise vector. For data fidelity items, and , This is a binary mask matrix, where the value is 1 at the sensor location and 0 in other areas. The generated data output by the generator. For high-resolution matrices, measured values ​​are only filled in at sensor locations. This indicates element-wise multiplication;

[0054] The training objective of the generator is to minimize the comprehensive loss function: ,in, To balance the hyperparameters.

[0055] As a preferred embodiment, the causal analysis and knowledge base module, when identifying the causal relationships of faults and storing fault modes, specifically includes:

[0056] S51: Spatiotemporal data collection and preprocessing, obtaining spatiotemporal sequence data of key node mechanical parameters from multi-source sensor network and parametric finite element simulation module to form observation dataset;

[0057] S52: Causal Structure Learning, based on the observation dataset, uses a causal discovery algorithm to learn the causal relationships between node state variables and construct a causal directed acyclic graph representing the causal dependencies between variables;

[0058] S53: Fault case simulation generation. Using the parametric finite element simulation module, various fault disturbance parameters are actively applied to simulate and generate spatiotemporal data of fault cases and their evolution process.

[0059] S54: Fault propagation pattern extraction. Based on the generated fault case data, analyze the entire process of each fault from its origin and evolution to its final destruction, and abstractly extract the fault propagation pattern that represents the fault propagation path and sequence.

[0060] S55: Knowledge base construction and updating, which integrates and correlates cause-effect graphs and fault propagation diagrams to build a fault knowledge base and provide reasoning basis for risk warning and hypothesis exploration modules.

[0061] Preferably, the probability analysis and early warning module, when performing probabilistic mechanical analysis and outputting risk probabilities, specifically includes:

[0062] S61: Uncertainty quantification input, which replaces the key input parameters in the parameterized finite element simulation module with their uncertainty measures instead of fixed values. The uncertainty measures are described in the form of a probability distribution and constitute the input random vector.

[0063] S62: Probabilistic finite element solution. Through Monte Carlo simulation, the parameterized finite element simulation module is sampled and solved multiple times to obtain the probability distribution of key response quantities.

[0064] S63: Probability interval calculation: Based on the probability distribution of the response quantity, calculate its predicted probability interval at a certain confidence level.

[0065] S64: Probabilistic risk assessment compares the actual data collected by real-time sensors with the predicted probability interval, calculates the tail probability of the data falling outside the interval, determines the risk level based on the tail probability value, and decides whether to trigger an early warning; the formula for calculating the tail probability is:

[0066] ;

[0067] in, Let R be the tail probability, and R be the response quantity. Let R be the cumulative distribution function of the response quantity. This refers to the actual data collected by the sensors in real time.

[0068] Specifically, when obtaining the probability distribution of key response quantities by performing multiple sampling solutions on the parametric finite element simulation module through Monte Carlo simulation, the process includes:

[0069] S71: Draw a set of samples from the distribution of the input random vector;

[0070] S72: Input the extracted sample as a deterministic parameter into the parameterized finite element simulation module, perform a deterministic finite element analysis, and obtain the response result;

[0071] S73: Repeat steps S71 and S72 to obtain a sample set of response quantities;

[0072] S74: Perform statistics on the sample set of response quantities and fit its empirical distribution function to approximate the true probability distribution of the response quantities.

[0073] Preferably, the active detection module, when in operation, specifically includes:

[0074] S81: Risk hypothesis generation. Based on the current real-time monitoring data, the fault propagation pattern stored in the causal analysis and knowledge base module, and the risk probability output by the probability analysis and early warning module, multiple potential fault risk hypotheses are generated.

[0075] S82: Sensing strategy planning, for each fault risk hypothesis, determines the key monitoring area and sensitive physical quantity most likely to show abnormal data when it is true, and generates a targeted sensor sampling instruction accordingly.

[0076] S83: Active perception verification sends sampling instructions to the adaptive sampling control module, interrupts its current strategy, temporarily increases the sampling frequency and accuracy of the sensor in the target area, and obtains high-fidelity verification data;

[0077] S84: Hypothesis Evaluation and Update. The obtained validation data is compared with the expected data patterns of the risk assumptions to calculate the probability of the assumptions being true. The system status and fault knowledge base is then updated based on the evaluation results. The probability of the assumptions being true is calculated using the following formula:

[0078] ;

[0079] in, Assuming failure risk, The posterior probability represents the updated probability that the proposed risk hypothesis is true, given the observed number of validations. Assumption The prior probability is determined by the initial confidence level. To verify the data, Let be the likelihood function, representing the probability of observing the current validation data under the assumption that the failure risk hypothesis is true.

[0080] Preferably, a data correction module is also included, used to establish a statistical or physical influence model between sensor data, and to use this model to perform cross-validation, denoising, and correction on abnormal sensor data, as well as to verify the rationality of the output results of the parametric finite element simulation module. The workflow is as follows:

[0081] S91: Impact Model Construction. Based on physical principles and historical monitoring data, an impact model is established between sensor data. The impact model includes a physical relationship model based on mechanical principles and a data-driven statistical relationship model.

[0082] S92: Real-time data verification. Substitute the real-time sensor data into the influence model, calculate the residual between the model's predicted value and the actual measured value. If the residual exceeds the preset tolerance range, mark the data as potentially abnormal data.

[0083] S93: Data correction and reconstruction. For data marked as potential anomalies, the corrected values ​​are calculated using the correlation data provided by the impact model and the state estimation algorithm.

[0084] S94: Simulation result verification. Substitute the output of the parameterized finite element simulation module into the influence model to verify whether it violates the known physical and statistical constraints. If it does, send a request to the model update module to re-perform parameter inversion.

[0085] A method for real-time monitoring and early warning of stress in a formwork system based on multi-source sensing includes the following steps:

[0086] S101: Collect physical parameters of the nodes in the formwork system through a sensor network;

[0087] S102: Update the parametric finite element simulation model using real-time data;

[0088] S103: Dynamically adjust the sampling strategy of each sensor based on intelligent algorithms;

[0089] S104: Generate full-field mechanical state information based on sparse data;

[0090] S105: Identify the causal relationships of failures and build a failure mode knowledge base;

[0091] S106: Perform probabilistic mechanical analysis and output the risk probability;

[0092] S107: Propose risk hypotheses and control sensors for active verification.

[0093] The beneficial effects of this invention are:

[0094] 1. Compared with the rigid alarm mechanism based on fixed thresholds commonly used in existing technologies, which is essentially a post-event response mode and cannot quantify the risk level, this solution introduces the probabilistic finite element analysis method, which treats input parameters such as material properties and load conditions as probability distributions. Through Monte Carlo simulation, the probability distribution interval of key mechanical responses is generated. Finally, the risk level is accurately quantified by calculating the tail probability of the measured data falling outside the probability interval. This early warning mechanism based on probability statistics not only significantly reduces the false alarm rate, but also achieves a fundamental leap from simple alarm to risk level assessment.

[0095] 2. Compared with the existing digital twin models that generally adopt the method of fixing the initial parameters and running, which is difficult to adapt to the dynamic characteristics of material performance evolution and environmental load changes during construction, this solution constructs a closed-loop parameter inversion calibration mechanism. By establishing a weighted deviation objective function between sensor measured data and simulation output, an intelligent optimization algorithm is used to automatically invert and update key model parameters such as material parameters and boundary conditions, so that the simulation model has the ability to continuously self-evolve and ensure that the digital twin maintains a high degree of dynamic consistency with the physical entity throughout its entire life cycle.

[0096] 3. Compared with existing monitoring systems that use fixed sampling strategies, which suffer from high energy consumption or blind spots, this solution is based on reinforcement learning to build an adaptive sampling mechanism. By dynamically adjusting the sampling frequency and accuracy of each sensor, it significantly reduces system energy consumption while ensuring risk identification capabilities, thus achieving optimal allocation of monitoring resources.

[0097] 4. Compared with the shortcomings of traditional monitoring systems that can only alarm but cannot analyze the fault mechanism, this solution combines causal discovery algorithms and fault simulation technology to build a knowledge base that includes causal graphs and fault propagation diagrams. It can identify the root cause of the fault from the monitoring data and predict the evolution path, providing in-depth decision support for proactive prevention and control.

[0098] 5. Compared with the existing technology's passive monitoring mode of waiting for anomalies to occur, this solution proposes a hypothesis-driven active detection mechanism. It generates risk hypotheses based on multi-source information and actively improves the sampling accuracy of the target area. The hypothesis probability is verified through Bayesian updates, realizing the transformation from passive response to active exploration.

[0099] 6. Compared with the limitations of traditional data correction methods that only process a single sensor in isolation, this scheme establishes a physical-statistical hybrid correlation model between multi-source sensor data. It includes both deterministic physical relationships based on mechanical principles and statistical correlation characteristics based on historical data. Through optimal estimation algorithms such as Kalman filtering, it achieves collaborative correction of multi-source data, which significantly improves the system's anti-interference ability and data reliability.

[0100] 7. In contrast to the model drift problem caused by the independent operation of the simulation model and the monitoring system in the existing technology, this solution designs a simulation-monitoring two-way verification mechanism. By influencing the model, the simulation results are physically reasonable and statistically consistent. For deviations exceeding the threshold, the model recalibration process is automatically triggered, forming the self-verification and self-correction capability of the digital twin system. Attached Figure Description

[0101] Figure 1 The diagram shown is a schematic representation of the structure of the real-time force monitoring and early warning system for the formwork system based on multi-source sensing according to the present invention.

[0102] Figure 2 The diagram shown is a flowchart of the real-time monitoring and early warning method for the stress of the formwork system based on multi-source sensing according to the present invention. Detailed Implementation

[0103] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0104] Please see Figures 1-2 The present invention provides an embodiment: a real-time force monitoring and early warning system for a formwork system based on multi-source sensing, comprising:

[0105] A multi-source sensor network deployed at key nodes of the formwork support system;

[0106] Multi-source sensor networks include:

[0107] Multiple physical quantity sensors are deployed on key stress nodes and potential failure areas of the support system, as determined by parametric finite element simulation analysis, to monitor the multidimensional mechanical state of the nodes in real time. The physical quantity sensors include at least:

[0108] Pressure sensors are used to monitor the axial pressure on vertical support members or nodes;

[0109] Strain sensors are attached to the surface of horizontal templates or beam components to monitor their bending strain;

[0110] Displacement sensors are deployed between the top and bottom of the support system or at key nodes to monitor the vertical settlement and horizontal displacement of the system.

[0111] Tilt sensors, mounted on independent support rods or the overall frame, are used to monitor the tilt angle.

[0112] Furthermore, the physical quantity sensor is connected to a data acquisition and transmission module via wired or wireless means to form a distributed sensor network. The data acquisition and transmission module is used to perform preliminary processing, aggregation, and uploading of sensor data to the computational data processing and calculation module.

[0113] Data acquisition and transmission module connected to sensor network;

[0114] The data acquisition and transmission module is directly connected to the multi-source sensor network and is responsible for powering, issuing commands, and acquiring data from various sensors distributed at key nodes of the support system. This module has a built-in signal conditioning circuit and analog-to-digital converter, which can filter, amplify, and convert the acquired analog signals into digital signals. It can also package and upload the aggregated multi-source sensor data to the cloud or edge computing server through its wired or wireless communication interface. At the same time, this module can receive and execute commands from the data processing and computing module to achieve remote dynamic configuration of the sensor sampling frequency and accuracy.

[0115] The data processing and calculation module, which communicates with the data acquisition and transmission module, specifically includes:

[0116] The parametric finite element simulation module is used to simulate the mechanical behavior of the formwork system. The specific workflow is as follows:

[0117] Model construction: Based on the geometric topology, material property parameter set, load condition parameter set and boundary condition parameter set of the support system, a parameterized finite element model is established. Among them, the parameters in the material property parameter set, load condition parameter set and boundary condition parameter set are dynamically adjustable variables.

[0118] Element calculation involves iterating through each element in the finite element model and calculating the element stiffness matrix for each element based on its type, geometric properties, and current material property parameters, using finite element theory. The underlying formula is as follows: ;in, The element stiffness matrix, The strain-displacement matrix, This is a matrix transpose operation. The elastic matrix is ​​used to characterize the constitutive relation of a material. Unit volume;

[0119] The overall stiffness matrix assembly involves assembling the element stiffness matrices of all elements into the overall stiffness matrix of the system based on the global number of their respective nodes.

[0120] Load vector assembly: Based on the current load condition parameter set, the overall load vector of the system is assembled.

[0121] Solve the equilibrium equations, specifically the system equilibrium equations composed of the global stiffness matrix, global displacement vector, and global load vector, to obtain the displacement solutions for all nodes. The relationship between strain and displacement at a point within an element is expressed using a strain-displacement matrix: ;in, For the strain at a point within the unit, This is the displacement vector of all nodes in this element;

[0122] The results are derived by calculating and outputting the stress, strain, and other derived mechanical responses of each element and node based on the obtained nodal displacement vectors, element shape functions, and material constitutive relations. The stress at a point within an element is calculated using constitutive relations. ;in, It is the elasticity matrix.

[0123] Specifically, this parametric finite element simulation module accurately simulates the mechanical behavior of a formwork system through a highly automated process. Its work begins with constructing a parametric model based on geometric topology, material properties, loads, and boundary conditions, all of which are dynamically adjustable variables. The core calculation process includes: integrating the strain-displacement matrix and the material elasticity matrix to solve for the stiffness matrix of each element; assembling all element stiffness matrices into the overall system stiffness matrix; simultaneously assembling the overall load vector formed by the load parameters; and then solving the system equilibrium equations to obtain the global displacement solution. Finally, based on the displacement solution, the module calculates and outputs the stress, strain, and other mechanical response results for the entire field element by element using the strain-displacement relationship and material constitutive relations, providing a simulation benchmark for subsequent real-time comparison and early warning.

[0124] The model update module is used to update the simulation model using real-time sensor data. Its specific workflow is as follows:

[0125] Define a parameter set and an objective function. Select key parameters to be inverted from the parametric finite element simulation module to form a vector, and define an objective function based on the deviation between the sensor measured data and the simulation data. The expression of the objective function is as follows: ;in, Let be the objective function. For the measured data of the i-th sensor, This refers to the mechanical response output of the simulation model, corresponding to the sensor's location and type, under the current parameter vector. Let n be the vector of parameters to be inverted, and n be the number of sensors. The importance weight of the i-th sensor data;

[0126] Iterative optimization is used to find the optimal solution by employing an optimization algorithm that minimizes the objective function. The parameter vector is automatically adjusted iteratively within the feasible region of the parameters.

[0127] The model is updated by feeding back the obtained optimal parameter solution to the parametric finite element simulation module, replacing the original parameters, and completing a calibration update of the simulation model.

[0128] Among them, when using an optimization algorithm to find the optimal solution by automatically iterating and adjusting the parameter vector within the feasible region of the parameters with the criterion of minimizing the objective function, the optimization algorithm used is one of gradient descent, quasi-Newton method and genetic algorithm.

[0129] Specifically, the model update module achieves dynamic online calibration of the simulation model through parameter inversion technology. Its core process first defines the key parameter vector to be inverted and constructs a target function with the weighted sum of squared deviations between the sensor measured data and the corresponding simulated output data. Then, an optimization algorithm (gradient descent, quasi-Newton method, or genetic algorithm) is used to automatically iteratively adjust the parameters to minimize the target function and find the optimal parameter solution that best approximates the measured data in the simulation results. Finally, the optimal parameters are fed back to the finite element simulation module to replace the original parameters, thereby completing the model calibration and ensuring that the digital twin and the physical entity maintain consistency in mechanical behavior.

[0130] The adaptive sampling control module dynamically adjusts the sampling strategies of each sensor. It incorporates a reinforcement learning agent. The agent's state space is based on the current health status and force complexity indicators of the support system, while its action space contains instructions for adjusting the sampling frequency and accuracy of each sensor. The reward function is a comprehensive evaluation of risk identification capability and total system energy consumption. The specific workflow of the adaptive sampling control module is as follows:

[0131] State perception: acquire and process real-time data from a multi-source sensor network and analysis results from a parameterized finite element simulation module to construct the state vector of the reinforcement learning agent.

[0132] In policy decision-making, the agent outputs an action vector based on its policy network according to the current state. The output action vector is an instruction to adjust the sampling frequency and sampling accuracy of each sensor in the network for the next time window.

[0133] Action execution involves sending action vectors to the data acquisition and communication module to dynamically configure the sampling parameters of each sensor.

[0134] Reward calculation involves calculating the reward based on the new state after a certain time period. The reward is used to evaluate the quality of the action. The underlying formula is as follows: ;in, For instant rewards, For the reward function, This is a risk identification reward function, whose value is positively correlated with the degree of abnormal risk discovered in the new state. This is an energy consumption penalty function, the value of which is positively correlated with the increase in the total power consumption of the system after the action is performed. To reinforce the learning agent's state vector, To reinforce the learning agent's action vectors, For risk identification reward weighting coefficient, This is the energy consumption penalty weighting coefficient;

[0135] The policy update involves storing the state transition sequence in the experience replay buffer and periodically sampling data to train and update the policy network in order to optimize long-term cumulative rewards.

[0136] The components of the state vector of a reinforcement learning agent include:

[0137] Real-time health status indicators are calculated directly from sensor data, including the system's maximum stress value, maximum displacement value, and rate of change.

[0138] The stress complexity index is calculated from the output of the parametric finite element simulation module, including the stress concentration factor and the variance or entropy value of the mechanical data between different nodes.

[0139] The current energy consumption index is the total system power consumption calculated from the power consumption model of each sensor in its current operating state.

[0140] The specific calculation method of the risk identification reward function is as follows: when the probability analysis and early warning module issues a risk warning based on the newly collected data, a positive reward is given according to the magnitude of the risk probability value of the warning. The reward value is a monotonically increasing function of the risk probability value of the warning. If no warning is triggered, the reward is zero or a small constant value.

[0141] The reinforcement learning agent is trained using an algorithm based on the Actor-Critic framework, which includes a policy network and a value network V or Q.

[0142] The adaptive sampling control module receives instructions from the active detection module. When the active detection module proposes a risky hypothesis, the adaptive sampling control module interrupts the current strategy of the reinforcement learning agent and directly sends action instructions with the highest sampling frequency and accuracy to the relevant sensors for active verification.

[0143] Among them, the parameterized finite element simulation module is used to assist in judging sampling requirements. Specifically, in the simulation results, regions with high stress, high displacement gradient or high theoretical failure probability are assigned higher sampling requirement weights. These weights are used as prior knowledge to initialize or constrain the policy learning process of the reinforcement learning agent.

[0144] Specifically, this adaptive sampling control module dynamically optimizes the sensor sampling strategy through a built-in reinforcement learning agent. The agent takes a state vector containing health status indicators, force complexity indicators, and system energy consumption as input, and outputs action commands to adjust the sampling frequency and accuracy of each sensor. The module operates continuously through a closed-loop process including state perception, policy decision-making, action execution, reward calculation, and policy update. Its reward function... (Risk Identification) (Energy Consumption) This module effectively balances the contradiction between risk monitoring accuracy and system energy consumption. It can receive instructions from the active detection module for the highest priority sampling, and at the same time use finite element simulation results as prior knowledge to guide the learning process, ultimately achieving the goal of allocating sensing resources on demand and intelligently optimizing system performance.

[0145] The data augmentation module is used to generate full-field mechanical state information based on sparse sensor data. The specific workflow is as follows:

[0146] Data preparation and preprocessing: Acquire sparse sensor measured data and corresponding full-field mechanical field data output by the parameterized finite element simulation module; normalize the data; and construct a training dataset.

[0147] Model training uses sensor data as a condition and the corresponding simulated full-field data as the target to train a generative AI model. Its goal is to learn the mapping relationship from sensor data to simulated full-field data.

[0148] High-resolution data generation involves inputting real-time sparse sensor data into a pre-trained generative AI model during online system operation to generate high-resolution full-field mechanical state data. The values ​​of the full-field mechanical state data at the sensor locations are consistent with the real-time sparse sensor data, and physically reasonable interpolation and extrapolation values ​​are generated in other regions.

[0149] The generative AI model is implemented using a robust generative adversarial network, including:

[0150] The generator takes random noise vectors and sparse sensor measured data as inputs to generate high-resolution full-field mechanical state data.

[0151] The discriminator takes the generated data output by the generator and the simulation data output by the parameterized finite element simulation module as inputs, and uses the measured data from the sparse sensor as a condition to determine the authenticity of the input data.

[0152] The generator and discriminator are trained adversarially using the following objective function:

[0153] ;

[0154] in, Let E be the objective function, where E represents the expected value, x is the actual full-field data sample, and y is the conditional data. For discriminator, Let z be the generator, and z be a random noise vector. For data fidelity items, and , This is a binary mask matrix, where the value is 1 at the sensor location and 0 in other areas. The generated data output by the generator. For high-resolution matrices, measured values ​​are only filled in at sensor locations. This indicates element-wise multiplication;

[0155] The training objective of the generator is to minimize the comprehensive loss function: ,in, To balance the hyperparameters.

[0156] Specifically, this data augmentation module utilizes a conditional generative adversarial network to generate high-resolution full-field mechanical state data from sparse sensor data. The module first trains a generative AI model using historical sensor data and corresponding finite element simulation full-field data. The generator takes sparse sensor data and random noise as input to generate full-field data, while the discriminator uses an adversarial loss function to determine the authenticity of the generated data compared to the simulation data. A data fidelity loss function is specifically introduced during training, forcing the generator's output values ​​at sensor locations to strictly match the actual measured values. In other areas, interpolation and extrapolation are performed based on physical laws, ultimately reconstructing physically plausible and detailed full-field stress and strain distribution cloud maps from limited measurement point data.

[0157] The causal analysis and knowledge base module is used to identify the causal relationships of faults and store fault modes. The specific workflow is as follows:

[0158] Spatiotemporal data collection and preprocessing: Spatiotemporal sequence data of mechanical parameters of key nodes are obtained from multi-source sensor network and parameterized finite element simulation module to form observation dataset;

[0159] Causal structure learning, based on the observation dataset, uses a causal discovery algorithm to learn the causal relationships between node state variables and construct a causal directed acyclic graph representing the causal dependencies between variables;

[0160] Fault case simulation generation utilizes a parametric finite element simulation module to actively apply various fault disturbance parameters, simulating and generating spatiotemporal data of fault cases and their evolution process.

[0161] Fault propagation pattern extraction: Based on the generated fault case data, analyze the entire process of each fault from its origin and evolution to its final destruction, and abstractly extract fault propagation patterns that represent the fault propagation path and sequence.

[0162] The knowledge base construction and updating integrates and correlates causal graphs and fault propagation diagrams to build a fault knowledge base and provide reasoning basis for risk warning and hypothesis exploration modules.

[0163] The causal discovery algorithm used is the PC algorithm, whose core is to gradually eliminate irrelevant edges and determine the direction of the edges based on the conditional independence test. For variables X and Y, the criterion for the conditional independence test is: given the variable set Z, whether the probability that X and Y are independent exceeds the predetermined significance level.

[0164] The fault knowledge base is also used for case-based reasoning. When online monitoring data matches an early pattern of a fault propagation pattern in the knowledge base, the knowledge base can predict its possible subsequent propagation path and final consequences, and push the prediction result to the risk warning and hypothesis exploration module.

[0165] Specifically, this causal analysis and knowledge base module constructs an intelligent fault knowledge base by integrating causal discovery and fault simulation technologies. The module first learns the causal relationships between node state variables from spatiotemporal monitoring data using a PC algorithm and constructs a causal directed acyclic graph. Simultaneously, it actively generates a large amount of fault case data through parametric finite element simulation, and then abstracts and extracts fault propagation patterns with clear propagation paths and sequence characteristics. Finally, the causal graph and fault propagation patterns are merged and stored to form a knowledge base. When online monitoring data matches early characteristics of fault modes in the database, subsequent propagation paths and consequences can be predicted, providing in-depth reasoning basis for risk warning.

[0166] The probability analysis and early warning module is used to perform probabilistic mechanical analysis and output risk probabilities. Its specific workflow is as follows:

[0167] Uncertainty quantification input replaces the key input parameters in the parametric finite element simulation module with their uncertainty measures instead of fixed values. The uncertainty measures are described in the form of a probability distribution and constitute the input random vector.

[0168] The probabilistic finite element method is used to obtain the probability distribution of key response quantities by performing multiple sampling and solving of the parametric finite element simulation module through Monte Carlo simulation.

[0169] Probability interval calculation: Based on the probability distribution of the response quantity, calculate its predicted probability interval at a certain confidence level.

[0170] Probabilistic risk assessment compares real-time sensor data with predicted probability intervals and calculates the tail probability of the data falling outside the interval. The risk level is determined based on the tail probability value, and a warning is triggered accordingly. The formula for calculating the tail probability is as follows: ;in, Let R be the tail probability, and R be the response quantity. Let R be the cumulative distribution function of the response quantity. This refers to the actual data collected by the sensors in real time.

[0171] The input random vector follows a specific probability distribution, and the uncertainty of the material's elastic modulus and load parameters is usually described by a log-normal distribution, with the probability density function being:

[0172] ;

[0173] in, Let be the probability density function. For the specific values ​​that the random variable can take, The location parameters of a log-normal distribution is the scaling parameter of the log-normal distribution.

[0174] Specifically, when obtaining the probability distribution of key response quantities by performing multiple sampling solutions on the parametric finite element simulation module through Monte Carlo simulation, the process includes:

[0175] Draw a set of samples from the distribution of the input random vector;

[0176] The extracted samples are used as deterministic parameters to input into the parameterized finite element simulation module, and a deterministic finite element analysis is performed to obtain the response results.

[0177] Repeat the previous two steps to obtain a sample set of response quantities;

[0178] Statistical analysis is performed on the sample set of response quantities to fit its empirical distribution function, approximating the true probability distribution of the response quantities.

[0179] Specifically, this probabilistic analysis and early warning module achieves a paradigm shift from deterministic early warning to probabilistic risk assessment through the probabilistic finite element method. The module first replaces deterministic values ​​for input quantities such as material parameters and loads with probability distributions (e.g., using a log-normal distribution to describe the elastic modulus). It then performs extensive sampling calculations on the parametric finite element model using Monte Carlo simulation to obtain the probability distributions and confidence intervals of key mechanical response quantities. Finally, it calculates the tail probability value of real-time monitoring data falling outside the predicted probability interval. This is used to quantify the risk level, and when the probability exceeds a threshold, an early warning is triggered, thus achieving a precise early warning mechanism based on the probability of risk occurrence rather than whether the threshold is exceeded.

[0180] The active detection module is used to propose risk hypotheses and control sensors to verify them. Its specific workflow is as follows:

[0181] Risk hypothesis generation: Based on current real-time monitoring data, fault propagation patterns stored in the causal analysis and knowledge base module, and risk probabilities output by the probability analysis and early warning module, multiple potential fault risk hypotheses are generated.

[0182] The sensing strategy planning identifies the key monitoring areas and sensitive physical quantities most likely to exhibit abnormal data when each fault risk hypothesis is true, and generates a targeted sensor sampling command accordingly.

[0183] Active perception verification sends sampling instructions to the adaptive sampling control module, interrupting its current strategy and temporarily increasing the sampling frequency and accuracy of the sensor in the target area to obtain high-fidelity verification data;

[0184] Hypothesis evaluation and updating involves comparing the acquired validation data with the expected data patterns of the risk assumptions, calculating the probability of the assumptions being true, and updating the system status and fault knowledge base based on the evaluation results. The probability of the assumptions being true is calculated using the following formula: ;in, Assuming failure risk, The posterior probability represents the updated probability that the proposed risk hypothesis is true, given the observed number of validations. Assumption The prior probability is determined by the initial confidence level. To verify the data, Let be the likelihood function, representing the probability of observing the current validation data under the assumption that the failure risk hypothesis is true.

[0185] Where, the likelihood function The degree of match between the validation data and the expected pattern of the failure risk assumption is calculated. The formula for calculating the degree of match is:

[0186] ;

[0187] in, The expected data pattern to be collected when the fault risk assumption is true.

[0188] Specifically, this active detection module achieves proactive risk identification and verification through a hypothesis-driven proactive sensing mechanism. Based on real-time monitoring data, fault propagation patterns, and risk probabilities, the module generates potential fault hypotheses. For each hypothesis, it plans a specific sensing strategy and interrupts the regular sampling process, forcibly increasing the sampling accuracy of sensors in the target area to obtain verification data. Finally, it uses a Bayesian update formula (… ) Calculate the posterior probability of the hypothesis being true, where the likelihood function is calculated by verifying the degree of match between the data and the expected pattern ( Quantification enables a paradigm shift from passive monitoring to active exploration.

[0189] The data correction module is used to establish statistical or physical influence models between sensor data, and to use these models to perform cross-validation, denoising, and correction on abnormal sensor data, as well as to verify the rationality of the output results of the parametric finite element simulation module. The workflow is as follows:

[0190] Impact model construction: Based on physical principles and historical monitoring data, an impact model between sensor data is established. The impact model includes a physical relationship model based on mechanical principles and a data-driven statistical relationship model.

[0191] Real-time data verification involves substituting real-time sensor data into the impact model and calculating the residual between the model's predicted value and the actual measured value. If the residual exceeds the preset tolerance range, the data is marked as potentially abnormal data.

[0192] Data correction and reconstruction: For data marked as potential anomalies, the corrected values ​​are calculated using the correlation data provided by the impact model and the state estimation algorithm.

[0193] Simulation result verification involves substituting the output of the parameterized finite element simulation module into the influence model to verify whether it violates known physical and statistical constraints. If it does, a request to re-perform parameter inversion is sent to the model update module.

[0194] Among them, the physical relationship model based on mechanical principles is a functional relationship derived from mechanics of materials and structural mechanics, and its expression is:

[0195] ;

[0196] in, Let be the measurement value of the i-th sensor. Let be the theoretical predicted value of the measurement value of the j-th sensor. For model vector parameters, such as nodal pressure based on Hooke's Law. With nodal displacement The relationship can be modeled as ,in, This is the equivalent stiffness coefficient;

[0197] Data-driven statistical relationship models are established through linear regression or Gaussian process regression, and their general expression is:

[0198] ;

[0199] in, For the measurements from the other n sensors, These are regression coefficients, obtained by training on historical normal data. This is the random error term that follows a normal distribution.

[0200] The process of calculating the corrected value using the state estimation algorithm includes two steps: prediction and update.

[0201] Prediction steps: ;

[0202] Update steps: ;

[0203] in, Let be the prior state estimation vector, representing the system state predicted purely from the system dynamic model before the measurement value at the current time k arrives. This represents the optimal state estimate obtained at the previous time k-1 after incorporating all measurement information up to time k-1. This is the state estimate at time k. Let k be the sensor measurement value at time k. Here is the state transition matrix. For the observation matrix, The Kalman gain matrix is ​​calculated based on the model prediction error covariance and the measurement noise covariance.

[0204] In this process, when substituting the output of the parametric finite element simulation module into the influence model to verify whether it violates known physical and statistical constraints, the relative deviation between the simulation results and the predicted values ​​of the influence model is calculated. The principle formula is as follows:

[0205] If relative deviation If the result exceeds the preset threshold, the simulation result is determined to violate the constraints and needs to be recalibrated.

[0206] in, This is a relative deviation. For simulation output values, These are the model's predicted values.

[0207] Specifically, this data correction module achieves sensor data cleaning and simulation result verification by constructing a physical and statistical influence model. First, the module establishes a physical / statistical relationship model between sensors based on mechanical principles (such as Hooke's Law P=K·D) and historical data. It identifies abnormal data by calculating the residual between measured values ​​and model predictions, and uses state estimation algorithms such as Kalman filtering for data correction and reconstruction. Simultaneously, it verifies the rationality of the simulation results by calculating the relative deviation δ=||y_simulated-y_model|| / ||y_model|| between the simulation results and model predictions. Deviations exceeding a threshold trigger a model recalibration request, forming a closed-loop verification mechanism spanning both the data and model dimensions.

[0208] A method for real-time monitoring and early warning of stress in a formwork system based on multi-source sensing includes the following steps:

[0209] The physical parameters of the nodes in the formwork system are collected through a sensor network;

[0210] Update the parametric finite element simulation model using real-time data;

[0211] The sampling strategy of each sensor is dynamically adjusted based on intelligent algorithms;

[0212] Generate full-field mechanical state information based on sparse data;

[0213] Identify the causal relationships of failures and build a failure mode knowledge base;

[0214] Perform probabilistic mechanical analysis and output the risk probability;

[0215] Propose risk hypotheses and control sensors to perform active verification.

[0216] Specifically, this technical solution achieves a paradigm shift in formwork system monitoring from passive alarm to proactive early warning through the deep integration of multi-source sensing and digital twin technologies. Its beneficial effects are mainly reflected in: significantly improving model accuracy and risk assessment reliability through parameter inversion and probabilistic finite element analysis; optimizing system energy consumption while ensuring risk identification capabilities by utilizing reinforcement learning adaptive sampling and AI data generation technologies; and revealing the intrinsic mechanisms of faults and achieving proactive risk intervention through causal analysis and proactive detection mechanisms. Ultimately, it forms a closed-loop intelligent management and control system encompassing perception, modeling, analysis, decision-making, and verification, significantly improving construction safety and management efficiency.

[0217] Example 1: Monitoring of Hydraulic Climbing Formwork System for Core Tube of Super High-Rise Building

[0218] Application Scenario: The core tube of a 400-meter super high-rise building is constructed using a hydraulic climbing formwork system. This system has a complex structure and is subjected to various dynamic loads, including concrete lateral pressure, wind load, and construction live load, placing extremely high demands on the real-time performance, accuracy, and reliability of the monitoring system.

[0219] Specific implementation process:

[0220] 1. System Deployment and Initial Modeling:

[0221] Sensor deployment: Pressure sensors (monitoring hydraulic cylinder pressure), strain gauges (monitoring strain of the frame chords), displacement sensors (monitoring the horizontal displacement of the frame relative to the structure), and tilt sensors (monitoring the overall verticality of the frame) were densely deployed at key load-bearing members, wall supports, and lifting points of the climbing formwork, as determined by the parametric finite element simulation module, forming a multi-source sensor network.

[0222] Initial model establishment: Based on the accurate drawings and design parameters of the climbing formwork system, a refined model including material properties (elastic modulus of steel E, Poisson's ratio v), loads (concrete pressure, wind load) and boundary conditions (wall-attached constraints) was established in the parametric finite element simulation module.

[0223] 2. System Operation and Intelligent Control:

[0224] Data Acquisition and Model Update: During the concrete pouring process, the data acquisition module uploads sensor data in real time. The model update module is automatically activated, using the measured displacement and strain data as the target, and employing a genetic algorithm to invert and calibrate the key load parameter of "concrete lateral pressure" in the simulation model, so that the simulation model quickly conforms to the actual working conditions.

[0225] Adaptive Sampling: The reinforcement learning agent in the adaptive sampling control module begins operation. Its state integrates the currently monitored maximum strain value (health state), the variance of different hydraulic cylinder pressure values ​​(stress complexity), and the total system power consumption (energy consumption). After making a decision, the agent issues a "highest precision sampling" command to the sensors in stress concentration areas, while reducing the sampling frequency to save energy in areas with stable stress.

[0226] Full-field state generation and causal analysis: Based on sparse sensor data, the data augmentation module uses a pre-trained conditional generative adversarial network to generate a high-resolution stress cloud map of the entire climbing formwork, showing stress concentration on one side of the top of the formwork. The causal analysis and knowledge base module analyzes historical data and finds a strong causal relationship (Granger cause) between the stress at this location and the displacement of a wall support below.

[0227] 3. Risk warning and proactive verification:

[0228] Probabilistic Early Warning: The probability analysis and early warning module sets input parameters such as material strength (yield strength) and load to a probability distribution (e.g., following a log-normal distribution), and calculates the probability distribution of failure in stress concentration areas through Monte Carlo simulation. If it finds that the current measured stress value has fallen outside the 99% confidence interval, with a tail probability as high as 0.15, the system immediately triggers a high-risk early warning.

[0229] Active Detection: Based on the aforementioned early warning and causal analysis, the active detection module generates a risk hypothesis: "Hypothesis H1: There is a risk of failure in the lower wall-mounted support, leading to load redistribution and stress concentration at the top." The module immediately sends a command to the adaptive sampling control module to interrupt the current strategy and force all sensors in the suspected failed support and its surrounding area to sample at the highest frequency and accuracy.

[0230] Hypothesis Verification and Decision Making: High-precision data verification revealed abnormal micro-movements in the support displacement. The data correction module used a physical model of the relationship between support displacement and pressure to cross-validate and denoise the data, confirming its authenticity. The system ultimately confirmed the hypothesis with a probability of 92%, clearly indicating the risk location and cause on the interface, guiding technicians to carry out emergency reinforcement, and successfully averting a potential major safety accident.

[0231] Beneficial effects: This embodiment demonstrates how the system can achieve closed-loop management of the entire process from "global perception" to "model calibration", then to "precise early warning" and "active verification" through the integration of multiple technologies under complex dynamic load environments, which greatly improves the safety control capabilities of super high-rise construction.

[0232] Example 2: Monitoring of the formwork support system for a large-span concrete roof of a large convention center

[0233] Application Scenario: A large convention center uses a cast-in-place concrete roof with a large span. Its formwork support system is classified as a high-risk and critical project. The support system has a large span and many members, posing a risk of stability failure. Therefore, the system needs to be able to perform comprehensive monitoring and overall stability assessment of large areas and multiple parameters.

[0234] Specific implementation process:

[0235] 1. System Deployment and Knowledge Base Construction:

[0236] Sensor Deployment: Within the tens of thousands of square meters of support area, a sensor network was deployed in zones and layers based on key locations determined by finite element analysis. Pressure sensors were deployed at the bottom of the uprights, strain gauges were attached to the horizontal connecting rods and scissor braces, and displacement sensors were deployed at the grid nodes at the top of the support system, forming a distributed monitoring network.

[0237] Fault Knowledge Base Pre-training: Before construction, various fault conditions were actively simulated using the parametric finite element simulation module, such as "single pole failure," "local uneven settlement of the foundation," and "missing horizontal connecting rods." The causal analysis and knowledge base module extracted the "propagation patterns" of various faults from these simulation data. For example, after pole failure, the path sequence of how the load is transmitted to adjacent members and triggers a chain reaction was constructed, thus building a rich fault knowledge base.

[0238] 2. System Operation and Intelligent Analysis:

[0239] Sparse Data and Global Perception: Due to the vast area, the sensor deployment is relatively sparse. The data augmentation module works continuously to generate displacement contour maps and pressure distribution maps covering the entire roof support system from limited point displacement and pressure data, enabling managers to intuitively grasp the stress and deformation status of the entire system.

[0240] Continuous learning and sampling optimization: The adaptive sampling control module runs continuously, and its reward function... The (risk identification) item remained a small constant value (because no obvious risks were found). Therefore, the agent gradually learned a sampling strategy that prioritized energy saving. Without affecting the overall monitoring, it significantly reduced the system's energy consumption, proving its value for long-term deployment.

[0241] 3. Identification and handling of hidden risks:

[0242] Causal Detection and Early Warning: Through the PC algorithm of the causal analysis module, the system discovered an abnormal statistical causal relationship between the pressure values ​​of multiple poles in a certain area. The pattern highly matched the early characteristics of the fault propagation pattern of "local foundation softening" in the knowledge base. Although none of the individual data points exceeded the threshold, the system still issued an early warning based on this correlation pattern.

[0243] Hypothesis-driven verification: The active detection module then generates the hypothesis "Hypothesis H2: The foundation in area A is softening, leading to a redistribution of forces between the poles." The module immediately instructs the sensors in that area to perform high-density sampling, and specifically increases the monitoring frequency of the foundation soil pressure sensors.

[0244] Data Correction and Model Validation: The data correction module, using a physical relationship model of pole pressure and displacement, discovered a small but systematic deviation (δ=3.5%) between the measured data and the model predictions, triggering a recalibration request for the simulation model. The model update module initiated inversion, ultimately identifying a 20% decrease in the "equivalent support stiffness" parameter of the foundation in this area.

[0245] Decision-making and response: The system, integrating all the above information, identified the hidden risks and precisely located the softened areas. Following the system's guidance, technicians reinforced the foundation in this area with grout, eliminating the potential risk of large-scale support system instability at its earliest stage.

[0246] Beneficial Effects: This embodiment demonstrates the advantages of the system in large-area, multi-parameter monitoring scenarios. It not only achieves "seeing the big picture from small details" through data augmentation, but also discovers hidden, systemic disease risks concealed in data correlations through causal analysis and a knowledge base. This enables "prevention-oriented" health management based on fault mechanisms, showcasing its high level of intelligence.

[0247] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A real-time stress monitoring and early warning system for a formwork system based on multi-source sensing, characterized in that: include: A multi-source sensor network deployed at key nodes of the formwork support system; Data acquisition and transmission module connected to sensor network; The data processing and calculation module, which communicates with the data acquisition and transmission module, specifically includes: The system comprises the following modules: a parametric finite element simulation module for simulating the mechanical behavior of the formwork system; a model update module for updating the simulation model using real-time sensor data; an adaptive sampling control module for dynamically adjusting the sampling strategies of each sensor; a data augmentation module for generating full-field mechanical state information based on sparse sensor data; a causal analysis and knowledge base module for identifying fault causal relationships and storing fault modes; a probabilistic analysis and early warning module for performing probabilistic mechanical analysis and outputting risk probabilities; and an active detection module for proposing risk hypotheses and controlling sensors for verification. The parametric finite element simulation module, when simulating the mechanical behavior of the formwork system, specifically includes: S11: Based on the geometric topology, material property parameter set, load condition parameter set, and boundary condition parameter set of the formwork system, establish a parameterized finite element model; S12: Calculate the element stiffness matrix of each element based on finite element theory; S13: Assemble the element stiffness matrices of all elements into the overall stiffness matrix of the system according to the global number of their respective nodes; S14: Assemble the overall system load vector based on the current load condition parameter set; S15: Solve the system equilibrium equations consisting of the overall stiffness matrix, the overall displacement vector, and the overall load vector to obtain the displacement solutions for all nodes; S16: Calculate and output the stress, strain and other derived mechanical response quantities for each element and node; The causal analysis and knowledge base module, in identifying the causal relationships of faults and storing fault modes, specifically includes: S51: Obtain spatiotemporal sequence data of key node mechanical parameters from the multi-source sensor network and parametric finite element simulation module to form an observation dataset; S52: Based on the observation dataset, a causal directed acyclic graph representing the causal dependencies between variables is constructed using a causal discovery algorithm. S53: Using the parametric finite element simulation module, various fault disturbance parameters are actively applied to simulate and generate spatiotemporal data of fault cases and their evolution process; S54: Extract the fault propagation pattern that represents the fault propagation path and sequence; S55: Integrate and correlate cause-effect graphs with fault propagation diagrams to construct a fault knowledge base; The probability analysis and early warning module, when performing probabilistic mechanical analysis and outputting risk probabilities, specifically includes: S61: Replace the key input parameters in the parametric finite element simulation module with their uncertainty metrics; S62: Through Monte Carlo simulation, the parameterized finite element simulation module is sampled and solved multiple times to obtain the probability distribution of key response quantities; S63: Calculate the predicted probability interval of the response quantity at a certain confidence level based on its probability distribution; S64: Compare the actual data collected by the real-time sensor with the predicted probability interval, calculate the tail probability that it falls outside the interval, determine the risk level based on the tail probability value, and decide whether to trigger an early warning.

2. The real-time force monitoring and early warning system for a formwork system based on multi-source sensing according to claim 1, characterized in that: When updating the simulation model using real-time sensor data, the model update module specifically includes: S21: Select the key parameters to be inverted from the parametric finite element simulation module to form a vector, and define an objective function based on the deviation between the sensor measured data and the simulation data; S22: An optimization algorithm is adopted, with minimizing the objective function as the criterion, and the parameter vector is automatically iteratively adjusted within the feasible region of the parameters to find the optimal solution; S23: Feedback the obtained optimal parameter solution back to the parameterized finite element simulation module to replace the original parameters.

3. The real-time force monitoring and early warning system for a formwork system based on multi-source sensing according to claim 2, characterized in that: The specific workflow of the adaptive sampling control module is as follows: S31: Construct the state vector of the reinforcement learning agent; S32: The agent outputs an action vector based on its policy network according to the current state; S33: Dynamically configure the sampling parameters of each sensor; S34: After a time period, calculate the reward based on the new state; S35: Store the state transition sequence in the experience replay buffer and periodically sample data to train and update the policy network.

4. The real-time force monitoring and early warning system for a formwork system based on multi-source sensing according to claim 3, characterized in that: When generating full-field mechanical state information based on sparse sensor data, the data augmentation module specifically includes: S41: Obtain sparse sensor measured data and corresponding full-field mechanical field data output by the parameterized finite element simulation module, normalize the data, and construct a training dataset. S42: Train a generative AI model using the corresponding simulated full-field data as the target. S43: Input sparse sensor data into a generative AI model to generate high-resolution full-field mechanical state data.

5. The real-time force monitoring and early warning system for a formwork system based on multi-source sensing according to claim 4, characterized in that: When the active detection module is working, it specifically includes: S81: Based on the current real-time monitoring data, the fault propagation pattern stored in the causal analysis and knowledge base module, and the risk probability output by the probability analysis and early warning module, generate multiple potential fault risk hypotheses; S82: For each fault risk hypothesis, determine the key monitoring area and sensitive physical quantity most likely to exhibit abnormal data when it is true, and generate a targeted sensor sampling instruction; S83: Send the sampling command to the adaptive sampling control module to obtain high-fidelity verification data; S84: Compare the acquired verification data with the expected data patterns of the risk assumptions, calculate the probability that the assumptions are true, and update the system status and fault knowledge base based on the evaluation results.

6. The real-time force monitoring and early warning system for a formwork system based on multi-source sensing according to claim 5, characterized in that: It also includes a data correction module, used to establish statistical or physical influence models between sensor data. The workflow is as follows: S91: Based on physical principles and historical monitoring data, establish an influence model between sensor data; S92: Substitute the real-time sensor data into the impact model, calculate the residual between the model's predicted value and the actual measured value, and mark it as potential abnormal data if the residual exceeds the preset tolerance range. S93: For data marked as potential anomalies, use the correlation data provided by the impact model and the state estimation algorithm to calculate the corrected value; S94: Substitute the output of the parametric finite element simulation module into the influence model to verify whether it violates known physical and statistical constraints.

7. A method for real-time monitoring and early warning of stress on a formwork system based on multi-source sensing, used to implement the real-time monitoring and early warning system for stress on a formwork system based on multi-source sensing as described in any one of claims 1-6, characterized in that: Includes the following steps: S101: Collect physical parameters of the nodes in the formwork system through a sensor network; S102: Update the parametric finite element simulation model using real-time data; S103: Dynamically adjust the sampling strategy of each sensor based on intelligent algorithms; S104: Generate full-field mechanical state information based on sparse data; S105: Identify the causal relationships of failures and build a failure mode knowledge base; S106: Perform probabilistic mechanical analysis and output the risk probability; S107: Propose risk hypotheses and control sensors for active verification.

Images