A concrete temperature monitoring method based on hydration heat self-power supply and temperature difference starting

Abstract

The application belongs to the technical field of temperature monitoring of nuclear island raft foundation mass concrete, and specifically discloses a concrete temperature monitoring method based on hydration heat self-power supply and temperature difference starting, which comprises the following steps: before concrete pouring, burying a passive self-powered temperature measuring element in a measuring point position determined based on a temperature field risk prediction result, the temperature measuring element being integrated with a thermoelectric conversion module, an energy storage module and a temperature sensing module; pouring concrete, using the heat released by the cement hydration reaction to form a temperature difference between the temperature sensing end and the reference end of the temperature measuring element; when the temperature difference reaches a preset starting threshold, the thermoelectric conversion module converts heat energy into electric energy and charges the energy storage module; after the voltage of the energy storage module reaches a working threshold, the temperature sensing module is powered to start and collect temperature data; and the collected temperature data is sent outward. The application can break away from the dependence on external power supply and realize intelligent self-starting precise monitoring.

Description

Technical Field

[0001] This application belongs to the field of temperature monitoring technology for large-volume concrete in nuclear island raft foundations, and more specifically, relates to a concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation. Background Technology

[0002] The raft foundation of the nuclear island, serving as the foundation of the reactor building in a nuclear power plant, is a typical large-volume concrete structure characterized by its large thickness (usually several meters), high strength grade, and large volume of concrete poured in a single operation. During cement hydration, a significant amount of heat is released, causing a rapid rise in the internal temperature of the concrete, while the surface dissipates heat more quickly, creating a temperature difference between the inside and outside. If this temperature difference is too large (typically controlled at 25°C), temperature cracks can easily occur, severely impacting the durability, integrity, and safety of the nuclear island foundation. Therefore, real-time and accurate temperature monitoring of the large-volume concrete of the raft foundation throughout its entire lifecycle is a crucial aspect of ensuring the quality of nuclear island construction.

[0003] Currently, the following technical means are mainly used for temperature monitoring of large-volume concrete in nuclear island raft foundations: First, pre-embedded thermocouple monitoring technology, which measures temperature by pre-embedding thermocouple sensors inside the concrete and using the thermoelectric effect, and connects to an external data acquisition instrument via an extension line for data recording and analysis; Second, fiber optic grating sensing monitoring technology, which measures temperature by pre-embedding fiber optic grating sensors inside the concrete and using the characteristic that the wavelength of the grating reflects changes with temperature, and transmits the signal to a demodulator via optical fiber for data acquisition.

[0004] However, existing technologies have the following shortcomings: Data acquisition devices typically require on-site power supply, which is extremely difficult and costly in nuclear island foundation pits with dense rebar mesh, overlapping construction machinery, and complex environments. Furthermore, the cables are easily damaged during construction, leading to measurement point failure. Long-distance leads between the temperature sensing element and the acquisition instrument not only increase material costs but also have numerous interfaces. Damage to any cable during concrete pouring and vibration means the entire measurement point is rendered unusable, resulting in poor system reliability. Monitoring equipment typically needs to be powered on and in standby mode before pouring, continuously consuming power. Manual intervention cannot guarantee real-time performance, easily leading to monitoring lag. After pouring, numerous pre-embedded cables protrude from the concrete surface, affecting subsequent construction and requiring trimming and protection, increasing workload. Measurement point layout schemes are mainly based on engineering experience and specifications, lacking scientific prediction of temperature field distribution under specific engineering conditions, potentially leading to insufficient monitoring in critical risk areas and redundant measurement points in other areas.

[0005] Therefore, how to break free from dependence on external power sources and achieve accurate monitoring of intelligent self-starting is an urgent problem to be solved. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this application is to provide a concrete temperature monitoring method based on self-powered hydration heat and temperature difference start-up, which can get rid of the dependence on external power source and realize intelligent self-starting accurate monitoring.

[0007] To achieve the above objectives, in a first aspect, this application provides a method for monitoring concrete temperature based on self-powered hydration heat and temperature difference activation, comprising the following steps: S10, Before concrete pouring, a self-powered temperature measuring element is embedded at the measuring point location determined based on the temperature field risk prediction results; the temperature measuring element integrates a thermoelectric conversion module, an energy storage module and a temperature sensing module. S20, pour concrete, and use the heat released by the cement hydration reaction to create a temperature difference between the sensing end and the reference end of the temperature measuring element. S30, when the temperature difference reaches the preset start-up threshold, the thermoelectric conversion module in the temperature sensing element converts thermal energy into electrical energy and charges the energy storage module; S40, when the voltage of the energy storage module in the temperature sensing element reaches the working threshold, power is supplied to the temperature sensing module to enable it to start and collect temperature data; S50 sends the collected temperature data outwards.

[0008] As a further preferred embodiment, in step S10, the location of the measuring point determined based on the temperature field risk prediction result is obtained by predicting the temperature field of concrete hydration heat through a physical information neural network. The training of the physical information neural network employs a composite loss function, which is defined as a weighted sum of data loss, physical loss, and boundary condition loss: ;in, To calculate the mean square error between the predicted and measured temperatures, This represents the mean square value of the residual term in the unsteady-state heat conduction equation. For boundary condition loss, and These are dynamically adjusted weighting coefficients; The residual term of the unsteady heat conduction equation R The calculation formula is: ,in T For temperature, t For time, The thermal conductivity varies with temperature. This is a heat source for cement hydration. The boundary condition is a third type of boundary condition, and its expression is: ,in k Thermal conductivity, Let be the normal derivative of temperature along the direction of the outward normal to the surface of the object. h The convective heat transfer coefficient is... The ambient temperature.

[0009] As a further preferred embodiment, the training of the physical information neural network employs a multi-stage progressive training strategy, including: Phase 1, Pure Data-Driven Pre-training: Setup , Using only data loss Pre-training was performed using the Adam optimizer with an initial learning rate of 1e. -3 Set the batch size to 64; The second stage involves the gradual introduction of physical constraints: employing a cosine annealing strategy to... Gradually increase from 0 to the target value of 0.1. The learning rate is gradually increased from 0 to a target value of 0.05, and a learning strategy is used to transition from a steady-state heat conduction scenario to a transient nonlinear scenario. The learning rate is decayed to 1e using cosine annealing. -4 ; Phase 3, Joint Optimization and Fine-tuning: Fixed =0.1、 With a target value of 0.05, the Adam optimizer is used in conjunction with a cosine annealing learning rate scheduler for joint optimization and fine-tuning. When the validation set... and If the number of training sessions does not decrease after 50 consecutive training rounds, the training will be terminated using an early stop mechanism.

[0010] As a further preferred embodiment, in step S10, the method for determining the location of the measuring point further includes: establishing a transient thermo-structural coupled finite element simulation model of the large-volume concrete of the nuclear island raft foundation, simulating the temperature field and stress field distribution during the concrete hydration process, and calculating the cracking risk coefficient. ,in Primarily tensile stress, To assess the tensile strength of concrete as it ages, areas with η ≥ 0.7 are identified as risk concern zones, and the layout of measuring points is optimized based on this assessment.

[0011] As a further preferred embodiment, in step S20, the reference end of the temperature sensing element is provided with a heat-insulating cavity, which is filled with silica aerogel felt or vacuum insulation board with a thermal conductivity of less than or equal to 0.02 W / (m·K) to slow down the direct conduction of heat from the sensing end to the reference end; the sensing end of the temperature sensing element is provided with a heat-conducting column, and the contact interface between the heat-conducting column and the concrete is filled with a high thermal conductivity potting compound or phase change thermally conductive material to reduce contact thermal resistance.

[0012] As a further preferred option, in step S30, a pulsed power supply strategy of storing energy first and then starting is adopted: the electrical energy generated by the thermoelectric conversion module in the temperature sensing element is first stored entirely in the energy storage module. After the energy accumulation reaches the working threshold, it provides instantaneous power to the temperature sensing module at once to complete a data acquisition task.

[0013] As a further preferred embodiment, in step S40, when collecting temperature data, an adaptive sampling frequency is adopted: the sampling and transmission frequency is dynamically adjusted according to the magnitude of the temperature difference or the rate of temperature change; during the concrete heating stage, data is collected once per hour; during the cooling stage, data is collected once per hour; however, when the cooling rate is greater than 2.0℃ / h or the total cooling rate is greater than 1.0℃ in 4 consecutive hours, data is collected once every 0.5 hours.

[0014] As a further preferred embodiment, the temperature sensing element also includes a microcontroller unit and an radio frequency module; in step S40, a graded startup logic is adopted: when the voltage of the energy storage module reaches a first threshold, the microcontroller unit enters the initialization state but shuts down the radio frequency transmission part; when the voltage continues to rise and reaches a second threshold higher than the first threshold, the microcontroller unit instructs the radio frequency module to transmit data; the microcontroller unit estimates the current temperature difference and energy input power based on the measured capacitor voltage rise rate, and decides the task to be executed this time upon power-up.

[0015] As a further preferred option, in step S50, the data is transmitted wirelessly, using frequency hopping spread spectrum technology. If no confirmation signal is received from the gateway, the channel is switched after a random delay for automatic retransmission. The temperature sensing element integrates a low-power wireless communication module, which sends a data packet containing the measurement point identifier and temperature value to an external receiving terminal. At the same time, the temperature sensing element also has a reserved wired data interface as a backup. When the wireless signal is blocked, data is collected and uploaded through a pre-embedded data bus.

[0016] Secondly, this application provides a self-powered temperature monitoring element for implementing the method described in any one of the above, comprising: Thermally conductive encapsulation housing for contact with the interior of concrete and conduction of heat; Thermoelectric conversion module, wherein the thermoelectric conversion module is a thermoelectric generator, the temperature sensing end of the thermoelectric generator is tightly thermally coupled to the thermally conductive encapsulation shell, and the reference end of the thermoelectric generator is provided with a heat insulation structure; An energy management and storage circuit, connected to the output of the thermoelectric conversion module, is used to boost, rectify, and manage the generated electrical energy, and store the processed electrical energy in the energy storage module; the energy management and storage circuit includes a maximum power point tracking circuit, used to dynamically adjust the input impedance to extract the maximum power from the thermoelectric conversion module; The temperature sensing module is connected to the energy management and storage circuit, and collects and processes temperature data after receiving power. A communication unit, connected to the temperature sensing module, is used to send the processed temperature data outward; the communication unit includes a low-power wireless communication module and a reserved wired communication interface, forming a dual-mode redundant communication mechanism.

[0017] The beneficial effects achieved by this application are as follows: (1) Zero dependence on external energy, achieving green construction: It completely eliminates the need for batteries or external power lines required by traditional monitoring solutions, and uses the waste heat generated during the cement hydration process as the only energy source, which is in line with the green and low-carbon construction concept. It is especially suitable for special occasions such as nuclear power plants with extremely high requirements for safety and cable laying and complex operating environments, eliminating the safety hazards and costs caused by wiring.

[0018] (2) Intelligent self-starting, realizing unmanned and accurate monitoring: The system innovatively uses a 5℃ temperature difference threshold as the "biological key" to intelligently identify the start and end of the concrete curing cycle and automatically turn on and off monitoring without any manual intervention. This not only avoids monitoring gaps caused by human negligence, but also achieves perfect synchronization between monitoring activities and construction progress.

[0019] (3) Extremely high system reliability and robustness: The system is in a "zero power consumption" state of complete power-off when there is no temperature difference, and there are no problems such as leakage corrosion of circuits and long-term standby loss of traditional batteries. At the same time, the wireless design reduces more than 90% of long-distance leads, which greatly reduces the risk of measurement point failure caused by collisions with construction machinery and vibration pulling.

[0020] (4) Flexible deployment of monitoring points greatly simplifies the construction process: Since the monitoring elements are freed from the constraints of cables, they can be arranged arbitrarily and flexibly with the steel mesh before pouring, just like "intelligent aggregates". They can even be buried in non-core areas by throwing during the pouring process. This greatly simplifies the pre-construction preparation work, shortens the construction period, and reduces the overall cost.

[0021] (5) Precise guidance for construction and improvement of project quality: By combining the simulation analysis in the early stage with the real-time monitoring data in the back end, a closed loop from prediction to verification is formed. The monitoring data can be used to verify the simulation model and provide valuable data support for temperature control and crack prevention in similar projects in the future, which fundamentally improves the construction quality of the nuclear island raft foundation, a key structure. Attached Figure Description

[0022] Figure 1 This is an overall flowchart of the method provided in the embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] This application provides a method for monitoring the temperature of large-volume concrete in nuclear island raft foundations without external power supply or human intervention. This method utilizes the hydration heat temperature difference formed after concrete pouring to power passive temperature measurement nodes through thermoelectric power generation technology, driving them to autonomously sense, collect, and transmit temperatures, thereby achieving a closed-loop monitoring scheme with zero external energy dependence, intelligent self-starting, and high reliability.

[0025] like Figure 1 As shown, the specific technical solution is as follows: Step 1: Predicting the temperature field and assessing crack risk in large-volume concrete based on Physical Information Neural Network (PINN) to guide the precise deployment of measuring points. To achieve precise allocation of monitoring resources and form a complete technical closed loop of "prediction-monitoring-verification-optimization", the following deep learning-based prediction and decision-making steps are performed before concrete pouring: 1.1 Construction of a Prediction Model for the Hydration Heat and Temperature Field of Large-Volume Concrete Integrating Physical Mechanisms and Historical Data This step abandons the purely data-driven "black box" model and constructs a physical information neural network. The physical equation of heat conduction is embedded as a constraint in the network loss function to ensure that the prediction results not only conform to the historical data patterns, but also strictly follow the basic laws of thermodynamics.

[0026] 1.1.1 Preprocessing and Feature Extraction of Multi-Source Heterogeneous Data Acquire and process the multi-source data required to build predictive models: Data Acquisition: Collect historical engineering data for the nuclear island raft foundation covering at least three different nuclear reactor types and more than five different construction seasons, including but not limited to: Material parameters: Cement hydration kinetic parameters (degree of hydration, adiabatic temperature rise rate), specific heat capacity, and thermal conductivity variation curves with age, derived from mix proportion tests; Geometric and construction parameters: 3D BIM model of nuclear island raft foundation, layer thickness, pouring sequence, pouring interval, and placement temperature; Environmental parameters: hourly air temperature, wind speed, and solar radiation intensity recorded by the meteorological station during the construction period; Measured data: Historical temperature data measured by pre-embedded thermocouples, and temperature crack detection reports (number, length, depth, and location distribution of cracks).

[0027] Physical consistency preprocessing: The first law of thermodynamics is used to verify the energy conservation of historical temperature data and eliminate outliers that clearly violate physical laws. The time-frequency domain features of temperature data are extracted by Fourier transform to identify the dominant frequency features of the hydration heat release cycle. Spatial clustering analysis was performed on the crack data to establish a database linking crack locations with temperature field characteristics (such as the location of the maximum temperature difference and the direction of the temperature gradient).

[0028] 1.1.2 Physical Information Neural Network (PINN) Architecture Design Network Structure: A multi-layer fully connected neural network architecture is adopted. The number of nodes in the input layer corresponds to the dimensions of the input parameters, including spatial coordinates (x, y, z), time t, and material parameters (cement type, fly ash content), construction parameters (entry temperature, pouring layer thickness), totaling 10-14 dimensions. The hidden layers consist of 5-6 layers, each with 96-128 neurons. The activation function chosen is the Swish function, which combines smoothness and non-linearity. The output layer is a single node, outputting the predicted temperature value. .

[0029] Physically constrained embedding: Calculating the partial derivatives of the network output with respect to input coordinates and time using automatic differentiation techniques. , , And so on, thus constructing the residual terms of the unsteady-state heat conduction equation:

[0030] in, The thermal conductivity varies with temperature. This is the heat source for cement hydration. This residual is treated as a physical loss. Introduce a total loss function.

[0031] Composite loss function design: The total loss function is defined as the weighted sum of data loss, physical loss, and boundary condition loss.

[0032] in: The mean square error between predicted temperature and measured temperature (and augmented data); The residual terms of the heat conduction equation force the model output to satisfy the physical laws of heat conduction. For boundary condition loss, the model is forced to satisfy specified heat dissipation boundary conditions (such as the third type of boundary condition) at the boundary coordinates:

[0033] Thermal conductivity represents a material's ability to conduct heat. It is the normal derivative of temperature along the direction of the outer normal to the surface of the object, that is, the temperature gradient in the direction perpendicular to the surface; The convective heat transfer coefficient characterizes the strength of heat transfer between the fluid and the solid surface.

[0034] The temperature of the surrounding fluid (environment); and The weight coefficients are dynamically adjusted. In the early stages of training, smaller values ​​are set to prioritize data fitting. As the training rounds increase, the values ​​are gradually increased so that the model gradually converges to a solution space that simultaneously satisfies physical laws and boundary conditions.

[0035] 1.1.3 Multi-stage training strategy for Physical Information Neural Network (PINN) To avoid training getting stuck in local optima and to ensure effective embedding of physical constraints, the following three-stage progressive training strategy is adopted: Phase 1: Pure Data-Driven Pre-training (0-800 rounds): Setup , , only use Training is then performed. This stage aims to allow the network to quickly learn the basic distribution characteristics of the data and establish a reasonable initial approximation of the temperature field. The Adam optimizer is used, with an initial learning rate set to 1e. -3 The batch size is set to 64.

[0036] Phase Two: Gradual Introduction of Physical Constraints (800-3000 rounds): Employing a cosine annealing strategy, Gradually increase from 0 to the target value of 0.1. The learning rate is gradually increased to 0.05. Simultaneously, a learning strategy is employed, starting with simple steady-state heat conduction scenarios to calculate physical losses, gradually transitioning to complex transient and nonlinear scenarios. During this stage, the balance between data fitting and adherence to physical laws is maintained, and the learning rate is decayed to 1e using cosine annealing. -4 .

[0037] Phase 3: Joint Optimization and Fine-tuning (after 3000 rounds): Fixed =0.1、 With a target value of 0.05, the Adam optimizer is used in conjunction with a cosine annealing learning rate scheduler for joint optimization. Monitoring the validation set... and If neither of the two decreases for 50 consecutive epochs, an early stopping mechanism is used to terminate training to prevent overfitting.

[0038] 1.1.4 Data Augmentation and Model Validation Data Augmentation: For regions with sparse historical data (such as construction under extreme temperature conditions or new mix proportions), a physics-guided data augmentation method is employed. Specifically, representative working conditions are simulated using classic finite element software (such as ANSYS, MIDAS, or ABAQUS), generating a batch of high-fidelity simulation data to supplement the training set, thereby enhancing the model's generalization ability in data-scarce regions. This simulation process strictly adheres to the physical equations of heat conduction to ensure the physical rationality of the augmented data.

[0039] Dataset partitioning: The collected historical engineering data and augmentation data were merged and divided into training, validation, and test sets in an 8:1:1 ratio. The training set was used for model parameter optimization, the validation set for hyperparameter tuning and early stopping, and the test set for independent performance evaluation of the final model.

[0040] Model validation: The model is comprehensively evaluated on an independent test set. Deterministic accuracy verification: Calculate the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) between the predicted and measured temperatures on the test set. The model should have an MAE ≤ 1.0℃ and an R² ≥ 0.95 on the test set.

[0041] Physical consistency verification: Randomly select new operating conditions in the test set that were not used in the training to verify whether the temperature field predicted by the model satisfies the first law of thermodynamics (i.e., the heat integral over the whole domain is in balance with the heat dissipation at the boundary), with an error requirement of ≤5%.

[0042] Uncertainty Calibration and Verification: For the prediction uncertainty of the Monte Carlo Dropout output in subsequent steps, a reliability curve is used for calibration and verification. Calculate the actual coverage probability of the predicted confidence interval (e.g., 90% confidence interval) to the true temperature value, requiring the deviation between the coverage probability and the nominal confidence level to be ≤5%, ensuring the reliability of uncertainty quantification.

[0043] 1.2 Model Uncertainty Quantification and High-Risk Area Identification Based on Monte Carlo Dropout To avoid the limitations of a single prediction result, this step introduces model uncertainty quantification technology to output temperature field predictions in probabilistic form, thereby more scientifically identifying risks.

[0044] Enable Monte Carlo Dropout: In the trained PINN model, retain the Dropout layer and ensure it remains effective during the prediction phase. Perform N (N≥100) forward propagations on the same input parameter (such as the coordinates at a certain moment), randomly discarding different neurons each time to obtain N predicted temperature values.

[0045] Generating the temperature field probability distribution: Based on N prediction results, the mean and variance of the temperature at each spatiotemporal point are calculated. The mean represents the most likely temperature value, while the variance quantifies the model's cognitive uncertainty in predicting at that point. Regions with high variance indicate that the model lacks confidence in the prediction results, which is usually caused by sparse training data or novel combinations of input parameters.

[0046] Definition of risk indicators based on integrated physical mechanisms: Risk Indicator 1: Probability of Maximum Inner-Surface Temperature Difference. Calculate the mean temperature difference between layers at different depths (e.g., the core area and the vicinity of the surface), and combine it with its variance to give the probability P (ΔT>25℃) that the temperature difference exceeds the 25℃ control threshold.

[0047] Risk Indicator 2: Temperature Gradient Vector Field. Calculate the mean and variance of the spatial temperature gradient. High gradient regions are key areas where temperature stress concentrates and cracks are prone to initiation.

[0048] Risk Indicator 3: Temperature Stress Concentration Factor Prediction. The temperature field (mean and variance) predicted by PINN is used as a thermal load and input into a fast finite element solver (or another trained stress PINN surrogate model) to directly output the distribution of tensile stress zones and the probability that the tensile stress exceeds the tensile strength of concrete.

[0049] 1.3 Optimization of Measurement Point Layout and Construction Scheme Decision Based on Information Entropy The results of quantifying the above uncertainties are transformed into specific action guidelines, achieving a leap from "experience-based deployment" to "information-driven deployment".

[0050] Active learning strategy for test point layout: Generate candidate measurement point set: Generate a large number of candidate measurement points within the 3D model of the raft foundation at a certain spatial resolution (e.g., 0.5m×0.5m×0.5m).

[0051] Calculating Information Gain: For each candidate monitoring point, the "information gain" that can be brought about by deploying a monitoring element at that point is evaluated using information entropy theory. Information gain consists of two parts: ① Risk-oriented: The higher the predicted risk value of the area where the point is located (such as the probability of temperature difference exceeding the threshold, the magnitude of tensile stress), the greater the information gain; ② Uncertainty-oriented: The higher the uncertainty (model prediction variance) at the point, the greater the information gain.

[0052] Optimal measurement point layout is generated by employing a greedy algorithm or Bayesian optimization to balance cost and effectiveness, automatically selecting a set of K measurement points that maximizes the total information gain. The final output is a "3D Coordinate Map of Optimal Measurement Point Layout," which not only marks the coordinates of the measurement points but also indicates the "key physical quantities" to be monitored at each point (such as core temperature, surface-to-internal temperature difference, and temperature gradient direction), as well as the "suggested installation orientation" of the monitoring elements (for example, for areas with the largest temperature gradient, it is recommended that the heat-conducting pillars of the elements be installed along the gradient direction to accurately capture the temperature difference).

[0053] Counterfactual optimization of the construction plan: Using a trained PINN model, counterfactual reasoning is performed. That is, with other parameters fixed, only one construction parameter is modified (such as adjusting the layer thickness from 1.0 meter to 0.8 meter), and the PINN model is rerun to quickly predict the probability distribution of the temperature field under the new scheme.

[0054] By comparing risk indicators (such as the probability of the maximum surface-to-internal temperature difference exceeding a threshold and the range of tensile stress zones) under different construction schemes, quantitative basis is provided for construction decisions. For example, if simulation shows that P(ΔT>25℃) = 85% under the original scheme, indicating an extremely high risk, but this probability drops to 15% after reducing the placement temperature, then the construction team is advised to adopt the latter scheme. This process transforms construction optimization from an "experience-based trial and error" process into a "data-driven, quantifiable, and predictable" intelligent decision-making process.

[0055] Step 2: Optimization of construction scheme and verification of measurement point scheme based on multiphysics finite element simulation As a preferred deployment method of this approach, for the same plane coordinate location (i.e., the same measurement position) with a large predicted temperature gradient and high risk of cracking, multiple temperature measuring elements can be set up and buried in layers along the concrete depth direction.

[0056] Based on the rapid PINN prediction in step 1, high-precision multiphysics finite element simulation analysis is carried out for key high-risk areas or new construction conditions to verify the PINN prediction results, optimize the construction plan, and check the rationality of the measurement point layout.

[0057] 2.1 Multi-source parameter integration and refined simulation model establishment Based on actual nuclear power engineering projects, a refined finite element simulation model of the large-volume concrete of the nuclear island raft foundation is established by integrating the following multi-dimensional parameters: 2.1.1 Construction of Concrete Thermodynamic Parameter Library Thermal parameters: The heat release curve of cement hydration was obtained through adiabatic temperature rise test, and the functional relationship between hydration degree and age was established; the nonlinear relationship between concrete specific heat capacity, thermal conductivity, and thermal conductivity with temperature was determined through thermophysical property test; the convective heat transfer coefficient between concrete surface and formwork and air was determined, taking into account the influence of formwork material (steel formwork / wooden formwork) and removal time.

[0058] Physical and mechanical parameters: The growth curve of the elastic modulus of concrete with age, creep coefficient, Poisson's ratio, and coefficient of linear expansion were measured; the change law of tensile strength of concrete with age was obtained through splitting tensile test, which serves as the basis for judging crack risk.

[0059] 2.1.2 Integration of Construction Conditions and Environmental Parameters Construction process parameters: According to the construction organization design, input the layer pouring thickness, the interval time between adjacent layers, the pouring sequence (such as symmetrical pouring, unidirectional advancement), vibration method and the range of influence.

[0060] Concrete pouring temperature control: The concrete pouring temperature is calculated using the heat balance method based on the temperature of the batching plant, transportation distance, ambient temperature and pouring time, taking into account temperature control measures for different seasons (such as adding ice in summer and heat preservation in winter).

[0061] Historical meteorological data: Hourly meteorological parameters during the construction period, including temperature, wind speed, wind direction, solar radiation intensity, and relative humidity, are acquired as boundary condition inputs. For continuous pouring processes lasting several days, a combination of actual weather forecasts and historical data from the same period is used for setting.

[0062] 2.1.3 Geometric Model and Mesh Generation Based on the nuclear island raft foundation design drawings, a full-size three-dimensional geometric model was established.

[0063] A hexahedral-dominated meshing strategy was adopted, with mesh refinement in areas where the temperature gradient was expected to be large (such as near the surface and around cooling water pipes) to ensure the accuracy of temperature and stress field calculations. The mesh size was controlled between 0.1m and 0.5m, and the total number of meshes was controlled between 100,000 and 200,000 depending on the model size.

[0064] 2.2 Transient Thermal-Structural Coupled Finite Element Analysis The following multiphysics coupling calculations were performed using finite element analysis software (such as ANSYS, ABAQUS, or MIDAS FEA): 2.2.1 Simulation of transient temperature field Based on the theory of unsteady heat conduction, with the heat balance equation as the core, and considering the heat of cement hydration as an internal heat source, the transient thermal analysis module is used to solve the spatiotemporal temperature distribution T(t, x, y, z) inside the concrete.

[0065] The hydration heat release model uses exponential or compound exponential functions to accurately simulate the variation of the hydration heat release rate with age.

[0066] Boundary condition settings: Apply third-type boundary conditions (convection + radiation) to the surface in contact with air; apply adiabatic or constant temperature boundaries to the bottom surface in contact with the foundation; apply adiabatic boundaries to the symmetry plane.

[0067] 2.2.2 Stress Field and Cracking Risk Assessment The temperature field calculation results are used as temperature loads and imported into the structural analysis module for thermal-structural coupling analysis.

[0068] The incremental iterative method is used to solve for the stress field distribution at each time step, with a focus on calculating the principal tensile stresses. and its relationship with the tensile strength of concrete The ratio of .

[0069] Define the cracking risk factor: When η ≥ 0.7, it is identified as a "risk concern area"; when η ≥ 1.0, it is identified as a "high-risk cracking area".

[0070] Considering the creep effect of concrete, the age-adjusted effective modulus (AEMM) method or Maxwell chain model is used to accurately simulate the mitigation effect of stress relaxation on cracking risk.

[0071] 2.2.3 Extraction of core predictive indicators Temperature field characteristic index: highest internal temperature of concrete and the time of occurrence; maximum surface-to-interior temperature difference Δ and its location; temperature gradient field Spatial distribution of T, identifying extreme regions of temperature gradient; cooling rate Identify areas that are cooling down too quickly.

[0072] Stress field characteristic index: maximum principal tensile stress Its spatiotemporal distribution; the extent of crack risk zone (the proportion of the area with η ≥ 0.7 in terms of volume); prediction of the location of potential crack initiation and the direction of crack propagation.

[0073] Construction sensitivity analysis: Through parametric scanning, the sensitivity of the impact of factors such as formwork temperature, layer thickness, and interval time on core indicators is analyzed, and key control parameters are identified.

[0074] 2.3 Verification of the fusion of simulation results and PINN predictions The PINN rapid prediction results from step 1 are cross-validated with the finite element detailed simulation results from step 2, forming a two-layer prediction system of "rapid screening-precise calculation": 2.3.1 Validation of Predictive Consistency Comparison of PINN prediction and finite element simulation in key performance indicators (PMI) Δ Deviation in high-risk areas. If the deviation is within the acceptable range for the project (e.g., If the deviation is ≤3℃ and the overlap of high-risk areas is ≥80%, then they corroborate each other and enhance the reliability of the prediction.

[0075] If the deviation exceeds the threshold, analyze the sources of the deviation: ① Insufficient representativeness of the data augmentation samples; ② Failure of the PINN model to converge during training; ③ Uncertainty in the input of finite element parameters. Optimize the model or parameters accordingly.

[0076] 2.3.2 Precise Positioning of High-Risk Areas The high-risk areas predicted by PINN (based on uncertainty quantification) are spatially superimposed with the high-risk cracking areas identified by finite element simulation (based on η ≥ 0.7).

[0077] Define the overall risk level: Level I Risk Zone: High-risk areas identified by both methods, where monitoring components are prioritized and a redundant deployment strategy is adopted.

[0078] Level II risk zone: Areas identified by only a single method, where monitoring elements are deployed proportionally based on engineering experience.

[0079] Level III risk area: Areas not identified by either method, where monitoring should be conducted at the minimum required level or with low density.

[0080] 2.4 Construction plan optimization and final confirmation of measurement point plan Based on simulation analysis and fusion verification results, the final construction guidance document is generated: 2.4.1 Suggestions for Optimizing the Construction Plan If simulation prediction or Δ Exceeding the control standards, a "Construction Plan Optimization Proposal" is generated, which includes: Adjustments to the layered pouring scheme: Recommend the optimal combination of layer thickness, interval time, and pouring sequence; Temperature control recommendations: Recommended mold entry temperature control target (e.g., ≤28℃ in summer), cooling water pipe layout parameters (spacing, flow rate, water flow duration), and surface insulation measures (insulation material thickness, covering time). Emergency response plan: Measures to deal with extreme weather (such as high temperature and cold wave), such as adjusting the pouring time, setting up curing sheds, and adding insulation layers.

[0081] Once the optimization plan is determined, return to step 1 and rerun the PINN model for rapid verification until the risk indicators meet the control requirements.

[0082] 2.4.2 Final confirmation of the measuring point layout plan Based on the combined information entropy optimization results from step 1 and the finite element cracking risk identification results from step 2, the final three-dimensional coordinate diagram of the measurement point layout is output.

[0083] In addition to the coordinates of the measuring points and the installation orientation from step 1, this drawing also includes the following information: Monitoring priority labeling: Level I risk areas are labeled "priority monitoring", Level II risk areas are labeled "routine monitoring", and Level III risk areas are labeled "reference monitoring"; Redundant deployment identification: Mark the measuring points in the Level I risk area with "dual component redundancy"; Component mounting depth correction: Based on the temperature gradient direction simulated by finite element method, the mounting angle of the heat-conducting pillar of the component is finely adjusted to ensure that it is consistent with the direction of the maximum temperature gradient, thereby maximizing the thermoelectric power generation efficiency. 2.4.3 Establishment of a Closed-Loop Mechanism for Simulation and Monitoring Data Establish a database linking simulation parameters and monitoring data. The real-time temperature data fed back by the monitoring system will be automatically compared with the simulation prediction values ​​at the corresponding measuring points.

[0084] When the deviation between the monitored value and the predicted value exceeds the set threshold (e.g., 5℃), the system triggers an early warning, prompting technicians to review the input parameters of the simulation model (e.g., whether the actual pouring temperature and environmental conditions are consistent with the prediction assumptions).

[0085] After the project is completed, all monitoring data will be archived into the historical database as high-value samples for subsequent incremental training of the PINN model, realizing a continuous learning loop of "prediction → monitoring → verification → optimization → re-prediction".

[0086] Step 3: Pre-installation and system deployment of self-powered temperature sensing elements based on simulation guidance Before the concrete pouring of the nuclear island raft foundation, self-powered temperature monitoring elements are fixed at key points inside the raft foundation (such as the center, upper middle, and edge areas) according to different thicknesses and heat dissipation conditions. The monitoring elements integrate a thermoelectric conversion module, an energy storage unit, a low-power temperature measurement chip, and a wireless transmission module. To achieve "plug-and-play" functionality and structural reuse, quick-connect and quick-release wire connectors are used between the temperature measurement elements and external data collection equipment. During construction, the temperature measurement elements are embedded along with the concrete, becoming consumables. Other structures (such as data repeaters, gateways, and other auxiliary equipment) are installed on the concrete surface and can be recycled after curing for reuse in subsequent projects.

[0087] 3.1 Refinement of Measurement Point Location: Instead of simply identifying "key points," a "3D Coordinate Map of Measurement Point Layout" is generated based on the simulation results from step 2. This map precisely marks the embedded coordinates (X, Y, Z) and installation orientation of each monitoring element. For example, for areas with the predicted maximum temperature gradient, the hot end of the element may need to be strictly oriented towards the center of the heat source; for edge areas, it is necessary to ensure that its cold end faces the template or air that may dissipate heat.

[0088] 3.2 Physical installation and protection of components: Specialized positioning fixture: To prevent component displacement during pouring and vibration, a plastic or metal positioning bracket that can be quickly snapped or tied to the reinforcing steel has been developed. This bracket not only secures the components but also allows adjustment of the component's embedding depth and angle according to the design drawings, ensuring that the orientation of its hot and cold ends meets expectations.

[0089] Redundant installation strategy: For critical measuring points such as the core area of ​​the structure, a dual-element redundant deployment can be adopted, that is, two independent elements are buried near the same coordinate (10-20cm apart). When one element fails unexpectedly, the other can still provide critical data.

[0090] Cable protection (for wired backup solutions): If wired backup is used, the outgoing cables should be protected with double-layer armor or flexible metal conduit, and securely fixed with cable ties at regular intervals along the main reinforcement bars of the structure to prevent them from being broken or flattened during concrete pouring. The ends of the cables should be uniformly numbered and collected in a waterproof junction box at the top of the foundation pit.

[0091] 3.3 System-level pre-configuration and calibration: ID-Location Mapping: Before installation, use a handheld terminal to scan the QR code or NFC tag on each component to read its unique ID. Then, using the coordinates on the construction drawings, associate and bind the ID with the physical location (e.g., "Unit 1 raft foundation, center point, depth -2.5m") in the background database in one go.

[0092] Communication network pre-configuration: For wireless solutions, the deployment location of the gateway needs to be planned in advance according to the site environment (such as high up on a tower crane, or fixed buildings around a foundation pit). At the same time, parameters such as channel, transmission power, and data reporting cycle (which can be designed as an adaptive algorithm that reports more frequently as the temperature difference increases) should be pre-configured for each monitoring element.

[0093] Factory calibration: All components must undergo temperature accuracy calibration before leaving the factory, and calibration coefficients must be provided. These coefficients are used for calibration during subsequent data processing to ensure the accuracy of the monitoring data.

[0094] Step 4: Engineering thermodynamic design for establishing and maintaining an effective temperature difference using the heat of hydration. A large volume of concrete is poured into the raft foundation. As the cement hydration reaction proceeds, the internal temperature of the concrete gradually rises. At this time, because one side of the monitoring element (the temperature-sensing side) is in close contact with the inside of the concrete, the temperature rises rapidly. The other side of the component (reference side) is relatively at a low temperature through a thermally conductive insulation structure or an exposed cold end. ).

[0095] 4.1 Thermal-structural co-design: The structure of the component itself is the guarantee for the establishment of temperature difference.

[0096] High-efficiency thermal coupling at the hot end: In addition to the thermally conductive shell, consider filling the interface between the component and the concrete with a high thermal conductivity potting compound or phase change thermally conductive material to eliminate micro-voids, minimize contact thermal resistance, and ensure that the hot end can respond instantly to temperature changes inside the concrete.

[0097] Thermal management design for the cold end: Miniature thermal insulation cavity: Inside the component, behind the cold end of the thermoelectric element, a closed miniature cavity is designed. The cavity is filled with silica aerogel felt or vacuum insulation board, with a thermal conductivity as low as 0.02 W / (m·K), which can greatly block the direct conduction of heat from the hot end to the cold end.

[0098] Cold end heat dissipation structure: If conditions permit (e.g., components are close to the concrete surface), miniature heat dissipation fins can be designed at the cold end to utilize the temperature difference between the high temperature inside the concrete and the external environment, and to further increase the temperature difference through heat radiation or weak convection to assist in heat dissipation.

[0099] Low thermal conductivity leads: All wires leading out from inside the component (such as wired backup antennas) must be made of extremely fine low thermal conductivity alloy wires covered with Teflon or stainless steel to prevent the wires from becoming "thermal bridges" for heat to flow from the hot end to the cold end.

[0100] 4.2 The dynamic evolution of temperature difference: Initial stage (0-12 hours after pouring): The internal temperature begins to rise, but the heat conduction is not yet stable. At this time, the cold end mainly relies on its thermal inertia to maintain the low temperature, the temperature difference increases rapidly, and the system enters the energy accumulation stage.

[0101] Peak period (12 hours to several days): The temperature difference reaches its maximum (possibly well over 5°C). This is the period when the system has the most abundant energy, enabling higher frequency data acquisition and transmission.

[0102] Decay period (several days later): As the heat of hydration weakens and internal heat is gradually conducted to the cold end, the temperature difference begins to shrink. The system's sampling frequency automatically decreases, and eventually, when the temperature difference falls below the threshold, it enters complete hibernation.

[0103] Step 5: Temperature Difference Start-up and Self-Power Supply When the temperature difference ΔT between the inside of the concrete and the reference side is greater than or equal to the activation threshold (5℃), the thermoelectric conversion module (semiconductor thermoelectric generator / TEG) built into the monitoring element begins to generate an electromotive force. This weak current is boosted and rectified by the power management circuit, and first charges the built-in energy storage capacitor. When the voltage of the energy storage capacitor reaches the operating voltage of the temperature sensing chip, the chip powers on and starts up.

[0104] 5.1 High-efficiency energy harvesting chain: Impedance matching: A maximum power point tracking (MPPT) circuit is designed between the thermoelectric generator and the boost converter. Because the output impedance of the generator changes with temperature variations, the MPPT dynamically adjusts the input impedance to ensure maximum power is extracted from the generator under any temperature difference.

[0105] Ultra-low startup voltage: Commercial energy harvesting chips with extremely low startup voltage (e.g., 20mV) are selected. This means that when the temperature difference is much less than 5°C, the chip begins to trickle charge the energy storage capacitor. When the temperature difference reaches or exceeds 5°C, the capacitor already has a certain amount of basic charge, which can reach the system startup voltage more quickly.

[0106] 5.2 Refined control of "storing energy first, then starting": Hierarchical startup logic: It's not simply a matter of "powering on when the voltage is reached." It can be designed so that when the capacitor voltage... When the first threshold V1 (e.g., 2.0V) is reached, the MCU enters the initialization state, but the RF section remains off; when Only when the voltage rises to a higher second threshold V2 (e.g., 2.5V) does the MCU instruct the RF module to transmit data. This tiered startup ensures sufficient power reserves for the most power-consuming stage: transmission.

[0107] Energy prediction and task scheduling: The MCU can predict the energy based on the measured rate of rise of the capacitor voltage ( It estimates the current temperature difference and energy input power, and intelligently decides the task to be performed upon power-up. For example, if... A larger temperature difference indicates a large temperature range and sufficient energy. In addition to the emission temperature data, a cold-end temperature measurement can also be performed. If the core temperature is relatively low, it will immediately go into hibernation after transmitting the core temperature data.

[0108] 5.3 Power supply redundancy: In addition to the supercapacitor, a miniature, non-rechargeable lithium thionyl chloride (Li-SOCl2) battery can be integrated as a final backup. Under normal circumstances, the system is powered by the capacitor; only in extreme cases (such as failure to establish a temperature difference) will the battery provide a one-time, extremely low-power "distress signal" to report the component status.

[0109] Step 6: Temperature Acquisition and Data Output After startup, the temperature sensing chip reads data from the built-in or external temperature sensor. In the case of a wired transmission scheme, the data is output through a pre-embedded redundant data line; in the case of a wireless scheme, the temperature data is sent to repeaters or receiving terminals around the pit through a low-power radio frequency module.

[0110] 6.1 Intelligent Data Acquisition: Adaptive sampling frequency: No longer a fixed interval. The MCU can dynamically adjust the sampling and transmission frequency based on the magnitude of the temperature difference or the rate of temperature change. For example, during the concrete heating phase, a sampling frequency of once per hour is used; after entering the plateau phase, it is once every 10 minutes; after entering the cooling phase, a sampling frequency of once per hour is used; however, when the cooling rate is greater than 2.0℃ / h or the total cooling rate over 4 consecutive hours is greater than 1.0℃, a sampling frequency of once every 0.5 hours is used. This allows for obtaining more data during critical periods while saving energy during non-critical periods.

[0111] Data validity assessment: The MCU can be equipped with a simple algorithm to assess the reasonableness of the collected temperature data, such as removing outliers that are significantly beyond physical limits due to transient interference, thereby improving data quality.

[0112] 6.2 Enhanced reliability of communication links: Multi-channel frequency hopping and automatic retransmission: Wireless communication uses frequency hopping spread spectrum technology to automatically switch between multiple channels to avoid interference. If no ACK confirmation signal is received from the gateway after transmission, the system will switch to another channel for automatic retransmission after a random delay to ensure that no data is lost.

[0113] Data packet simplification and compression: To reduce energy consumption per transmission, the data packet is designed to be extremely simplified. It only contains the component ID, temperature value (converted to integer), and Cyclic Redundancy Check (CRC) code, and no longer includes a timestamp (the timestamp is added uniformly by the gateway after receiving the data).

[0114] Redundant deployment of gateways: Deploy at least two backup wireless gateways around the foundation pit to ensure that data can still be received by the other gateway if one gateway fails or its signal is temporarily blocked.

[0115] Step 7: Sleep and Cycle After data transmission is complete, the system enters deep sleep mode, retaining only the extremely low-power voltage monitoring circuit. If the temperature difference persists and the capacitor voltage drops and then recharges, the system is awakened to perform the next data acquisition; if the temperature difference disappears (below 5°C), the system completely shuts down and returns to zero, waiting for the next temperature rise cycle.

[0116] 7.1 Deep Sleep Mode Design: Peripheral power cut-off: In sleep mode, the MCU completely cuts off the power supply to peripherals such as temperature sensors and RF chips through an ultra-low on-resistance load switch, ensuring zero power consumption for these modules.

[0117] The MCU's ultimate low-power mode: The MCU enters the deepest sleep mode, shutting down all unnecessary clocks and peripheral functions, retaining only an ultra-low-power comparator for monitoring capacitor voltage. The entire system's sleep current can be controlled below 1 microamp (μA).

[0118] Timed wake-up (optional): In addition to voltage wake-up, an extremely low-power timer (such as wake-up every 24 hours) can be configured to perform a "heartbeat" detection when there is no temperature difference or the temperature difference is very small, and send a very short data packet to inform the monitoring center that the measuring point is "online and in normal condition", but the power is insufficient or the temperature difference is insufficient.

[0119] 7.2 End-of-Lifecycle Management: Permanent hibernation mechanism: When the temperature difference remains below the threshold (e.g., <3℃) for more than a preset, long confirmation time (e.g., 72 hours), the system logic determines that the maintenance cycle has completely ended. Thereafter, even if the temperature difference briefly exceeds the threshold due to accidental temperature disturbances (e.g., exposure to sunlight), the system will ignore it or enter a stricter startup logic to avoid generating meaningless interference data during non-maintenance periods.

[0120] Data archiving and component "invisibility": After confirming that all monitoring points have entered permanent dormancy, the monitoring center archives the batch of monitoring data. These components embedded in concrete do not require recycling, and their completely passive and chemical-free properties ensure that they have no negative impact on the structure or the future environment, achieving physical "invisibility".

[0121] (III) Optimization of Technical Solutions Optimization setting of temperature difference start-up threshold: The start-up threshold is preferably set to 5℃. This value is determined based on the following considerations: 1) The temperature difference is much lower than the control temperature difference for the risk of concrete cracking (usually 25℃), ensuring that the system has been reliably started before the risk of cracking occurs; 2) The temperature difference is higher than the diurnal temperature difference of the normal construction environment (usually less than 5℃), which can effectively filter out false starts caused by ambient temperature fluctuations, ensuring that the system only works under the temperature difference dominated by the heat of hydration of concrete.

[0122] Energy storage unit configuration and energy management strategy: A pulsed energy management strategy of "storing energy first, then starting up" is adopted. Supercapacitors are preferred as the energy storage unit due to their advantages such as ultra-long cycle life (matching the construction cycle), no risk of chemical leakage, and high power density, making them ideal for short-term, high-power pulsed loads (such as wireless transmission). The system does not aim for real-time power supply, but rather wakes up and completes monitoring tasks all at once after sufficient energy has accumulated. This mechanism greatly lowers the barrier to utilizing weak and discontinuous energy sources.

[0123] Dual-mode communication backup mechanism: The monitoring element can simultaneously integrate a wireless transmission interface (such as LoRa) and a wired M-Bus / RS485 interface. Under normal operating conditions, wireless transmission is prioritized for minimal deployment. In extreme conditions (such as severe signal shielding from the dense multi-layered steel reinforcement mesh within the nuclear island raft foundation), the system can switch to wired mode, aggregating and uploading data via a pre-embedded redundant bus to ensure reliable communication.

[0124] Optimized component structure and packaging: The monitoring component is encased in a high-strength, high-thermal-conductivity engineering plastic or metal shell. One side of the shell features a raised "heat-conducting pillar" or a metal heat-conducting end cap to increase the contact area with concrete, enabling rapid and accurate heat sensing. The other side has a closed "insulation cavity" filled with super-insulating materials such as aerogel to maintain a relatively low cold-end temperature. The hot end of the thermoelectric generator is tightly fitted to the heat-conducting shell, while the cold end is connected to the heat dissipation structure within the insulation cavity, ensuring a stable temperature difference.

[0125] The key point of this application is: (1) PINN three-stage progressive training strategy Technical features: Phase 1: Pure data-driven pre-training → Phase 2: Gradual introduction of physical constraints → Phase 3: Joint optimization and fine-tuning, combined with course learning, cosine annealing, and early stopping mechanism.

[0126] Technical benefits: Avoids training from getting stuck in local optima, ensuring that the model has both data fitting accuracy and physical consistency.

[0127] (2) Hierarchical start-up logic and energy prediction scheduling Technical features: A first threshold (MCU initialization) and a second threshold (RF transmission) are set. The MCU estimates the temperature difference based on the capacitor voltage rise rate (dV / dt) and makes intelligent decisions on task execution strategies.

[0128] Technical benefits: Ensures sufficient power for the most power-consuming components and optimizes energy utilization efficiency.

[0129] (3) Adaptive sampling frequency Technical features: The MCU dynamically adjusts the sampling and transmission frequency based on the temperature difference or the rate of temperature change (high frequency during the heating phase, medium frequency during the plateau phase, and low frequency during the cooling phase).

[0130] Technical benefits: More data is obtained during critical periods, and energy is saved during non-critical periods.

[0131] (4) Dual-mode redundant communication mechanism Technical features: The component integrates both a wireless communication module (LoRa) and a wired communication interface (M-Bus / RS485), prioritizing wireless under normal operating conditions and switching to wired under extreme conditions.

[0132] Technical benefits: Ensures reliable communication and adapts to complex construction environments.

[0133] (5) Maximum Power Point Tracking (MPPT) Technical features: An MPPT circuit is designed between the thermoelectric generator and the boost converter to dynamically adjust the input impedance.

[0134] Technical effect: Ensures maximum power extraction under any temperature difference.

[0135] (6) Low-power deep sleep and lifecycle management Technical features: Power supply to peripherals is cut off during hibernation, hibernation current ≤1μA; after the temperature difference remains below the threshold (<3℃) for more than the confirmation time (72 hours), it enters permanent hibernation.

[0136] Technical benefits: Extreme energy saving, avoiding data interference during non-maintenance periods.

[0137] (7) High-efficiency thermal coupling design at the hot end Technical features: thermally conductive pillars / thermally conductive end caps + high thermal conductivity potting compound / phase change thermally conductive material filling interface.

[0138] Technical benefits: Eliminates microscopic gaps, reduces contact thermal resistance, and enables rapid response to temperature changes.

[0139] (8) Multi-channel frequency hopping and automatic repeater Technical features: The wireless communication uses frequency hopping spread spectrum technology. If no ACK is received, the channel is switched and retransmitted after a random delay.

[0140] Technical effect: Avoids interference and ensures no data loss.

[0141] (9) Redundant deployment of gateways Technical features: Deploy at least two wireless gateways around the foundation pit, which serve as backups for each other.

[0142] Technical effect: Data can still be received even if any gateway fails or the signal is blocked.

[0143] (10) Finite element simulation and PINN fusion verification Technical features: PINN rapid prediction is cross-validated with finite element fine simulation to define a comprehensive risk level (Level I / Level II / Level III).

[0144] Technical effect: It forms a two-layer prediction system of "rapid screening-precise calculation" to improve the reliability of prediction.

[0145] The following is a specific implementation example of this application: Example: Taking the temperature monitoring of the raft foundation of a nuclear island in a nuclear power plant as an example, the raft foundation is disc-shaped, with a diameter of approximately 50 meters and a thickness of 5.5 meters. The concrete strength grade is C40, and the volume of concrete poured in one continuous operation is approximately 8,000 cubic meters. The construction period is in August, with an average local temperature of 28°C and a diurnal temperature range of approximately 4°C.

[0146] Implementation steps: Step 1: Temperature field prediction and measurement point layout optimization based on Physical Information Neural Network (PINN) 1.1 Data Collection and Model Building (1) Historical data collection The technicians first collected the following historical engineering data: Material parameters: Concrete mix proportion test data from 6 domestic nuclear power projects (covering AP1000, Hualong One, EPR and other reactor types), including different varieties such as slag silicate cement and ordinary silicate cement, fly ash content ranging from 15% to 35%, and corresponding hydration heat release curves, specific heat capacity, thermal conductivity and other data as a function of age.

[0147] Geometric and construction parameters: BIM model of the nuclear island raft foundation for each project, layered pouring scheme (layer thickness of about 0.5m), pouring sequence, and temperature record of pouring into the formwork (20℃-28℃).

[0148] Environmental parameters: hourly temperature, wind speed, and solar radiation intensity recorded by the meteorological station during construction.

[0149] Measured data: approximately 2,000 sets of historical temperature data measured by the pre-embedded thermocouples, and crack detection reports after construction (including crack location, length, and depth).

[0150] (2) Physical consistency preprocessing The collected data was processed as follows: The energy conservation of temperature data was verified using the first law of thermodynamics, and 17 sets of abnormal data caused by sensor failure were eliminated. By extracting the time-frequency domain features of temperature data through Fourier transform, it was identified that the main frequency of the hydration heat release cycle is concentrated in 12-48 hours after pouring. Spatial clustering analysis was performed on the crack data to establish a database linking crack locations with temperature field characteristics. It was found that 85% of the cracks occurred in regions with a temperature gradient >15℃ / m.

[0151] (3) PINN model construction and training Constructing a physical information neural network: Network structure: 12-dimensional input layer (spatial coordinates x, y, z; time t; cement type code; fly ash content; pouring temperature; pouring layer thickness; ambient temperature; wind speed); 5 hidden layers, each with 128 neurons; activation function is Swish; output layer is the predicted temperature. .

[0152] Physical constraint embedding: embedding the residuals of the unsteady heat conduction equation As a physical loss item.

[0153] Multi-stage training: Phase 1 (0-800 rounds): Purely data-driven pre-training. It decreased from 0.52 to 0.18; Phase Two (800-3000 rounds): Gradually introduce physical constraints. From 0 to 0.1, It decreased from 0.23 to 0.08; Phase 3 (after 3000 rounds): Joint optimization and fine-tuning, stopping as early as round 4200.

[0154] Model validation: On the test set, MAE=1.6℃, R²=0.97, physical consistency error 3.8%, and uncertainty calibration bias 3.2%.

[0155] 1.2 Uncertainty Quantification and High-Risk Area Identification Enable Monte Carlo Dropout (N=200 forward propagations): Temperature field probability distribution: The mean and variance of temperature at each spatiotemporal point were calculated. The results show that the core region (depth 2.5-4.0m) has the highest mean temperature (predicted 68℃) and a smaller variance (±1.2℃); the near-surface region (depth 0-0.5m) has a larger temperature variance (±2.5℃), mainly due to the uncertainty of environmental conditions.

[0156] Risk indicator calculation: Probability of maximum surface temperature difference: The probability that the temperature difference between the core area and the surface exceeds 25℃ two days after the third layer is poured is calculated as P(ΔT>25℃)=85%; Temperature gradient vector field: The temperature gradient is found to be the largest at the boundary between the 2nd and 3rd layers (approximately 2.2m deep), reaching 18℃ / m; Temperature stress concentration factor: By inputting the temperature field into the stress PINN surrogate model, it is predicted that the tensile stress zone is mainly concentrated in the central region of the 2nd and 3rd layers and the edge region of the 4th layer.

[0157] 1.3 Optimization of measuring point layout and decision-making on construction plan (1) Active learning of measurement point layout Generate a candidate measurement point set: Generate approximately 10,000 candidate measurement points within the 3D model of the raft foundation at a resolution of 0.5m × 0.5m × 0.5m; Calculate information gain by combining risk-oriented (probability of temperature difference, magnitude of tensile stress) and uncertainty-oriented (prediction variance); Output the optimal solution: Using Bayesian optimization, under the constraint of 25 measurement points, the set of measurement points that maximizes the total information gain is automatically selected, generating a "3D coordinate diagram of the optimal measurement point layout," with annotations as follows: Measurement point A (core area, coordinates x=0, y=0, z=-3.0m): monitors the core temperature; it is recommended that the heat-conducting column be installed vertically upwards. Measuring point B (at 1 / 4 thickness, coordinates x=0, y=0, z=-1.2m): monitor the temperature gradient; it is recommended that the heat-conducting column be installed horizontally. Measuring point C (near the surface, coordinates x=0, y=0, z=-0.3m): monitors surface temperature; Measurement point D (edge ​​area, coordinates x=12.5m, y=0, z=-2.5m): monitors the heat dissipation boundary, it is recommended that the cold end face the template.

[0158] (2) Counterfactual optimization Using a trained PINN model for counterfactual reasoning: Original plan: 1.1m thick layer × 5 layers, no cooling water pipes → P(ΔT>25℃)=85%; Option 1: Adjust the layer thickness to 0.9m × 6 layers → P(ΔT>25℃)=62%; Option 2: Add cooling water pipes (1.5m spacing, water supply for 72 hours) → P(ΔT>25℃)=23%; Option 3: Option 2 + mold temperature reduced from 28℃ to 25℃ → P(ΔT>25℃)=12%.

[0159] Based on the simulation results, Scheme 3 is recommended, and a "Construction Scheme Optimization Proposal" is generated.

[0160] Step 2: Optimization of construction scheme and verification of measurement point scheme based on multiphysics finite element simulation 2.1 Multi-source parameter integration and refined simulation model establishment (1) Concrete thermo-mechanical parameter library The adiabatic temperature rise test showed that the peak value of the hydration heat release curve appeared 48 hours after casting, with a peak heat release rate of 12.5 J / (kg·h). Thermophysical property tests showed that the specific heat capacity was 0.96-1.05 kJ / (kg·K) and the thermal conductivity was 2.2-2.8 W / (m·K). Mechanical tests showed that the elastic modulus at 28 days was 32.5 GPa, the tensile strength was 3.2 MPa, and the creep coefficient was 2.1.

[0161] (2) Construction conditions and environmental parameters Construction process: 5 layers of pouring, each layer 1.1m high, with a 3-day interval between layers, and symmetrical pouring sequence; The temperature at which the material enters the mold is calculated to be 27.5℃ using the heat balance method (the temperature at the outlet of the mixing plant is 26℃, the temperature rise during transportation is 1.5℃, and the temperature drop during the pouring process is negligible). Meteorological data: Hourly temperatures in August (highest 35℃, lowest 24℃), wind speed 1.5-3.0 m / s; Cooling water pipes: set according to the optimized scheme, with a spacing of 1.5m×1.5m, a pipe diameter of 32mm, a water flow rate of 1.2 m³ / h, a water temperature of 20℃, and a water flow time of 72 hours.

[0162] (3) Geometric model and mesh generation A full-size 3D model was created, using a hexahedral dominant mesh with finer meshes near the surface. The mesh size was 0.1-0.5m, and the total number of meshes was approximately 200,000.

[0163] 2.2 Transient Thermal-Structural Coupled Finite Element Analysis (1) Simulation of transient temperature field Transient thermal analysis was performed using ANSYS, with a calculation time step of 1 hour and a total duration of 30 days. The results showed that the highest internal temperature occurred 48 hours after the third layer was poured, reaching 67.2℃; the maximum temperature difference between the inside and the surface occurred 52 hours after the third layer was poured, reaching 28.5℃ (67.2℃ in the core area and 38.7℃ on the surface).

[0164] (2) Stress field and cracking risk assessment The temperature field is introduced as a load into the structural analysis, taking into account the creep effect (AEMM model). Cracking risk coefficient η distribution: η=0.85-0.92 in the central area of ​​the 2nd and 3rd layers (risk concern area), η=0.78 in the edge area of ​​the 4th layer (risk concern area), and no high-risk area with η≥1.0 was found; Construction sensitivity analysis: For every 1°C decrease in the formwork temperature, It decreased by approximately 1.2°C.

[0165] 2.3 Verification of the fusion of simulation results and PINN predictions (1) Prediction Consistency Verification PINN Prediction =68.0℃, Finite Element Simulation =67.2℃, deviation 0.8℃; PINN predicts high-risk areas (centers of layers 2 and 3 and edge of layer 4), and finite element simulation identifies risk concern areas (η≥0.7) with an 88% overlap in location. The deviations are within an acceptable range and corroborate each other.

[0166] (2) Precise positioning of high-risk areas Spatial overlay analysis determines the overall risk level: Level I risk zone: The central area of ​​the 2nd and 3rd floors (within a radius of approximately 3m) should prioritize the deployment of monitoring components, using a dual-component redundant deployment. Level II risk zone: The edge area of ​​the 4th layer (ring-shaped area), arranged at a density of 2 per 10m²; Level III risk zone: Other areas shall be arranged in accordance with the minimum requirements of the regulations.

[0167] 2.4 Construction plan optimization and final confirmation of measurement point plan (1) Final confirmation of the measuring point layout plan Based on the PINN optimization results and finite element verification, the final 3D coordinate diagram of the measurement point layout is output, including: A total of 25 measurement points were set up: 6 in the Level I risk area (dual-element redundancy, 12 elements in practice), 8 in the Level II risk area, and 5 in the Level III risk area; Monitoring priority marking: Level I area monitoring points are marked "priority monitoring", and the data reporting frequency is set to 5 minutes / time; Redundant deployment identification: Level I zone measurement points are marked with "Dual component redundancy"; Component mounting depth correction: Based on the temperature gradient direction (mainly vertical), the heat-conducting pillars are installed vertically; for edge areas where the temperature gradient has a horizontal component, the heat-conducting pillars are installed at a 15° angle.

[0168] (2) Establishment of a closed-loop mechanism for simulation and monitoring data Establish a parameter association database to associate and bind the coordinates, component ID, PINN predicted value, and finite element simulation value of each measuring point.

[0169] Step 3: Prefabrication, pre-configuration, and field deployment of self-powered temperature sensing elements 3.1 Component Structure and Performance Parameters Self-powered temperature sensing element: Dimensions: Diameter 35mm, Height 28mm; Thermally conductive outer casing: copper alloy, thermal conductivity 380 W / (m·K); Thermoelectric generator: TEG1-127-1.0, maximum output power 1.2W (ΔT=50℃); Cold end insulation cavity: filled with silica aerogel, with a thermal conductivity of 0.018 W / (m·K); Supercapacitor: 1F / 5.4V, cycle life >100,000 cycles; MCU: STM32L0 series, sleep current 0.3μA; Wireless module: LoRa, frequency 470MHz, transmit power 10dBm; Temperature sensor: DS18B20, accuracy ±0.5℃.

[0170] 3.2 System Pre-configuration and Calibration (1) Factory calibration All 25 components (including 12 redundant ones, for a total of 37) undergo temperature accuracy calibration before leaving the factory. Five-point calibration was performed in the constant temperature bath (0℃, 20℃, 40℃, 60℃, 80℃). Each component obtains its calibration coefficients and writes them into the EEPROM; After calibration, the accuracy is ±0.3℃.

[0171] (2) ID-location mapping Use a handheld device to scan the QR code on each component and bind it to the backend database: Component ID: SHT-001 ~ SHT-037; Physical location: Correspond to the coordinates in the "Final Measurement Point Layout 3D Coordinate Map".

[0172] (3) Communication network pre-configuration Gateway deployment: The main gateway is installed at the top of the tower crane (40m high), and the backup gateway is installed in the temporary building on the east side of the foundation pit; Channel configuration: Use the 470-510MHz frequency band, automatic frequency hopping, and avoid the construction walkie-talkie channel; Reporting cycle: Adaptive algorithm, report every 5 minutes when temperature difference > 15℃, report every 15 minutes when temperature difference 5-15℃, and report every hour when temperature difference < 5℃.

[0173] 3.3 On-site deployment and installation (1) Locating and setting out measuring points Use a total station to accurately lay out the coordinates of the measuring points on the steel reinforcement mesh, with an error of ±2cm.

[0174] (2) Component fixing Uses a dedicated positioning bracket: Core area measuring points: Secure them between the double-layer steel mesh with cable ties to ensure the heat-conducting columns are vertically upward; Edge zone measuring points: fixed on the reinforcing bars near the formwork, with the cold end facing the formwork; Redundant deployment: For measurement points in the Level I risk zone, two independent components are buried 15cm apart near the same coordinates.

[0175] (3) Wired backup cable laying For measurement points in Level I risk areas, an M-Bus bus is reserved: The cable is protected by double-layer armor. Secure the main reinforcing bars with cable ties every 0.5m. The ends converge at the waterproof junction box at the top of the foundation pit, and are uniformly numbered.

[0176] Step 4: Concrete pouring and temperature difference establishment 4.1 Pouring process The concrete is poured continuously over a period of approximately 70 hours, with a placement temperature of approximately 26°C.

[0177] 4.2 Dynamic Evolution of Temperature Difference Eight hours after pouring (first layer), the temperature in the core area began to rise, with the hot end temperature reaching about 35°C, while the cold end remained at about 25°C due to thermal hysteresis, resulting in a temperature difference of 10°C. The system then began to enter the energy accumulation stage.

[0178] 48 hours after pouring (after the completion of the third layer), the core area temperature reached its peak. Measuring point A (core area): hot end temperature 67.5℃, cold end temperature 39.2℃, temperature difference 28.3℃; Measuring point B (at 1 / 4 thickness): hot end temperature 58.3℃, cold end temperature 41.5℃, temperature difference 16.8℃; Measuring point C (near the surface): hot end temperature 45.2℃, cold end temperature 32.8℃, temperature difference 12.4℃; Measuring point D (edge): hot end temperature 52.6℃, cold end temperature 29.4℃ (towards the template for heat dissipation), temperature difference 23.2℃.

[0179] 4.3 Verification of Thermal-Structure Co-design Thermal coupling at the hot end: The heat-conducting column is in close contact with the concrete, and the temperature response time is less than 30 seconds; Cold end insulation: The aerogel cavity effectively blocks heat conduction, and the temperature at the cold end is 8-12℃ lower than that of the surrounding concrete. Low thermal conductivity leads: Teflon-coated leads do not form obvious thermal bridges.

[0180] Step 5: Temperature Difference Start-up and Self-Power Supply 5.1 Energy Harvesting Start-up When the temperature difference ΔT reaches 5℃, the TEG starts to output voltage: When the temperature difference is 10℃, the TEG output voltage is approximately 0.8V; When the temperature difference is 20℃, the output voltage is approximately 1.6V; When the temperature difference is 28.3℃, the output voltage is approximately 2.2V.

[0181] 5.2 Energy Management Process (1) Trickle charging stage The power management chip (LTC3108) starts operating when the input voltage is ≥20mV. When the temperature difference is 5℃, the input voltage is about 0.4V, and the output voltage is boosted to 3.3V to charge the supercapacitor with a current of about 5mA. Approximately 8 minutes later, the capacitor voltage was charged to 2.0V (the first threshold), and the MCU entered the initialization state, while the RF section remained off.

[0182] (2) Staged start-up When the capacitor voltage reaches 2.0V: the MCU initializes, reads the temperature sensor, but does not transmit; Continue charging for about 5 minutes until the capacitor voltage reaches 2.5V (second threshold): The MCU instructs the RF module to transmit data.

[0183] (3) Energy prediction and task scheduling The MCU determines the energy status based on dV / dt: Peak period (dV / dt≈0.3V / min): Determine sufficient energy to execute the complete task (read temperature + emit); Decay period (dV / dt≈0.05V / min): When energy is deemed insufficient, the system immediately goes into hibernation after reaching core temperature.

[0184] 5.3 Power Supply Redundancy Verification Under normal operating conditions, the system is powered entirely by supercapacitors. In a laboratory simulation of a temperature difference failure scenario (temperature difference <3℃), the micro-battery (CR1220) successfully provided a one-time "distress signal" transmission.

[0185] Step 6: Temperature Acquisition and Data Output 6.1 Intelligent Data Acquisition (1) Adaptive sampling frequency The MCU dynamically adjusts based on the temperature difference: Days 2-4 (temperature difference > 15℃): Collect and transmit data every 5 minutes; Days 5-7 (temperature difference 8-15℃): Collect samples every 15 minutes; Days 8-14 (temperature difference 3-8℃): Samples were collected every 30 minutes; After 14 days (temperature difference <3℃): stop collecting data and enter permanent dormancy.

[0186] (2) Data validity judgment MCU built-in algorithm for judgment: A temperature change rate exceeding 5℃ / min is considered abnormal and the data is discarded. After three consecutive abnormalities, the measurement point is marked as a "suspected fault" and the reporting frequency is reduced.

[0187] 6.2 Communication Link Reliability (1) Data packet structure Each data packet contains only: Component ID: 4 bytes; Temperature value: 2 bytes (integer, resolution 0.1℃); CRC checksum: 2 bytes; Total length 8 bytes, transmission time <50ms.

[0188] (2) Multi-channel frequency hopping and automatic repeater It uses the 470-510MHz frequency band, with a total of 16 channels, and automatically hops frequencies every 5 seconds; If no ACK is received from the gateway, the channel will be switched and the data will be retransmitted after a random delay of 500-2000ms. It can be retransmitted up to 3 times with a success rate of 98.5%.

[0189] (3) Gateway redundancy When the primary gateway fails, the backup gateway automatically takes over, with a switching time of less than 1 second and no data loss.

[0190] 6.3 Data Transmission Examples Taking measurement point A (component ID: SHT-001) as an example: Time: 52 hours after pouring; Sampling temperature: 67.5℃; Data packet: [0x01, 0x00, 0x01, 0x00, 0x02, 0xA3, 0x00, 0x1C] (ID=1, temperature=675=67.5℃, CRC=0x001C); Transmit power: 10dBm; Receiving gateway: Main gateway (top of tower crane), signal strength -75dBm; Transmission delay: <100ms.

[0191] Step 7: Sleep and Cycle 7.1 Deep Sleep Mode After data transmission is complete: The MCU disconnects the power supply to the temperature sensor and RF chip via a load switch; The MCU enters Stop mode, retaining only the voltage monitoring comparator; Measured sleep current: 0.8μA.

[0192] 7.2 Wake-up and Loop When the supercapacitor discharges to below 2.0V, the system shuts down and enters hibernation mode. When the temperature difference persists, the TEG continues to charge the capacitor, and wakes it up again after the voltage rises back to 2.5V; During peak periods (temperature difference > 15℃), the device wakes up approximately 12 times per hour. During the decay period (temperature difference <8℃), the device wakes up approximately 2-3 times per hour.

[0193] 7.3 End of Life Cycle 30 days after pouring: Temperature measurement can be stopped when the temperature difference between the concrete surface and the ambient temperature is less than 20℃. All 37 components have entered permanent sleep mode and no longer respond to any temperature changes; After confirmation by the monitoring center, all monitoring data will be archived to the historical database; The components are permanently embedded in concrete, eliminating chemical pollution and achieving physical "invisibility".

[0194] Effect evaluation (1) Monitoring data integrity Throughout the entire monitoring period (30 days): A total of 2496 valid data points were collected: measuring point A, measuring point B, measuring point B, measuring point C, and measuring point D. Data integrity rate: 98.2% (data loss was mainly caused by temporary signal blockage, which was recovered after automatic retransmission); The lack of a measuring point caused the failure due to a power supply problem.

[0195] (2) Temperature evolution curve The monitoring data fully recorded the temperature evolution: Maximum internal temperature: 67.5℃ recorded at measuring point A (predicted 68.0℃, simulated 67.2℃); Maximum surface-to-interior temperature difference: 28.5℃ (predicted 29.1℃, simulated 28.8℃); Peak time: 52 hours after pouring (predicted 48-56 hours); Cooling rate: 0.8℃ / day on average after peak.

[0196] (3) Construction guidance effect Monitoring data showed that the maximum temperature difference on the third day had reached 24℃, which was close to the control threshold. The surface insulation was immediately strengthened on site (covered with two layers of cotton quilts). The temperature difference was controlled between 22-25℃ on the 4th and 5th days, and did not exceed the 25℃ control standard. Finally, non-destructive testing was conducted on the surface of the raft foundation, and no harmful temperature cracks were found.

[0197] (4) System autonomous operation verification It automatically starts when the temperature difference reaches 5℃, requiring no manual intervention; The sampling frequency is automatically increased during peak periods and automatically decreased during decay periods, resulting in no energy waste. After maintenance is completed, it will automatically and permanently enter a dormant state, requiring no manual recycling or power outage.

[0198] (5) Zero external energy dependence Throughout the entire monitoring period: No batteries or external power source were used; The only energy source is the heat of hydration of concrete; It has achieved intelligent monitoring that allows for "immediate use upon burial and immediate shutdown after use".

[0199] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A concrete temperature monitoring method based on hydration heat self-power supply and temperature difference start-up, characterized in that, Includes the following steps: S10, Before concrete pouring, a self-powered temperature measuring element is embedded at the measuring point location determined based on the temperature field risk prediction results; the temperature measuring element integrates a thermoelectric conversion module, an energy storage module and a temperature sensing module. S20, pour concrete, and use the heat released by the cement hydration reaction to create a temperature difference between the sensing end and the reference end of the temperature measuring element. S30, when the temperature difference reaches the preset start-up threshold, the thermoelectric conversion module in the temperature sensing element converts thermal energy into electrical energy and charges the energy storage module; S40, when the voltage of the energy storage module in the temperature sensing element reaches the working threshold, power is supplied to the temperature sensing module to enable it to start and collect temperature data; S50 sends the collected temperature data outwards.

2. The method for concrete temperature monitoring based on self-power supply by hydration heat and temperature difference initiation according to claim 1, characterized in that, In step S10, the location of the measuring point determined based on the temperature field risk prediction result is obtained by predicting the temperature field of concrete hydration heat through a physical information neural network. The training of the physical information neural network employs a composite loss function, which is defined as a weighted sum of data loss, physical loss, and boundary condition loss: ;in, To calculate the mean square error between the predicted and measured temperatures, This represents the mean square value of the residual term in the unsteady-state heat conduction equation. For boundary condition loss, and These are dynamically adjusted weighting coefficients; The residual term of the unsteady heat conduction equation R The calculation formula is: ,in T For temperature, t For time, The thermal conductivity varies with temperature. This is a heat source for cement hydration. The boundary condition is a third type of boundary condition, and its expression is: ,in k Thermal conductivity, Let be the normal derivative of temperature along the direction of the outward normal to the surface of the object. h The convective heat transfer coefficient is... The ambient temperature.

3. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 2, characterized in that, The training of the physical information neural network employs a multi-stage progressive training strategy, including: Phase 1, Pure Data-Driven Pre-training: Setup , Using only data loss Pre-training was performed using the Adam optimizer with an initial learning rate of 1e. -3 Set the batch size to 64; The second stage involves the gradual introduction of physical constraints: employing a cosine annealing strategy to... Gradually increase from 0 to the target value of 0.

1. The learning rate is gradually increased from 0 to a target value of 0.05, and a learning strategy is used to transition from a steady-state heat conduction scenario to a transient nonlinear scenario. The learning rate is decayed to 1e using cosine annealing. -4 ; Phase 3, Joint Optimization and Fine-tuning: Fixed =0.1、 With a target value of 0.05, the Adam optimizer is used in conjunction with a cosine annealing learning rate scheduler for joint optimization and fine-tuning. When the validation set... and If the number of training sessions does not decrease after 50 consecutive training rounds, the training will be terminated using an early stop mechanism.

4. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 1, characterized in that, In step S10, the method for determining the location of the measuring point further includes: establishing a transient thermo-structural coupled finite element simulation model of the large-volume concrete of the nuclear island raft foundation, simulating the temperature and stress field distribution during the concrete hydration process, and calculating the cracking risk coefficient. ,in Primarily tensile stress, To assess the tensile strength of concrete as it ages, areas with η ≥ 0.7 are identified as risk concern zones, and the layout of measuring points is optimized based on this assessment.

5. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 1, characterized in that, In step S20, the reference end of the temperature sensing element is provided with a heat-insulating cavity, which is filled with silica aerogel felt or vacuum insulation board with a thermal conductivity of less than or equal to 0.02 W / (m·K) to slow down the direct conduction of heat from the sensing end to the reference end; the sensing end of the temperature sensing element is provided with a heat-conducting column, and the interface between the heat-conducting column and the concrete is filled with a high thermal conductivity potting compound or phase change thermally conductive material to reduce contact thermal resistance.

6. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 1, characterized in that, In step S30, a pulsed power supply strategy of storing energy first and then starting is adopted: the electrical energy generated by the thermoelectric conversion module in the temperature sensing element is first stored in the energy storage module. After the energy accumulation reaches the working threshold, it provides instantaneous power to the temperature sensing module at once to complete a data acquisition task.

7. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 1, characterized in that, In step S40, when collecting temperature data, an adaptive sampling frequency is used: the sampling and transmission frequency is dynamically adjusted according to the magnitude of the temperature difference or the rate of temperature change; during the concrete heating stage, data is collected once per hour; during the cooling stage, data is collected once per hour; however, when the cooling rate is greater than 2.0℃ / h or the total cooling within 4 consecutive hours is greater than 1.0℃, data is collected once every 0.5 hours.

8. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 1, characterized in that, The temperature sensing element also includes a microcontroller unit and an RF module; in step S40, a graded startup logic is adopted: when the voltage of the energy storage module reaches a first threshold, the microcontroller unit enters the initialization state but shuts down the RF transmission part; when the voltage continues to rise and reaches a second threshold higher than the first threshold, the microcontroller unit instructs the RF module to transmit data; the microcontroller unit estimates the current temperature difference and energy input power based on the measured capacitor voltage rise rate, and decides the task to be executed this time upon power-up.

9. The concrete temperature monitoring method based on self-powered hydration heat and temperature difference activation as described in claim 1, characterized in that, In step S50, the data is transmitted wirelessly using frequency hopping spread spectrum technology. If no confirmation signal is received from the gateway, the channel is switched after a random delay for automatic retransmission. The temperature sensing element integrates a low-power wireless communication module, which sends a data packet containing the measurement point identifier and temperature value to an external receiving terminal. At the same time, the temperature sensing element also has a reserved wired data interface as a backup. When the wireless signal is blocked, data is collected and uploaded through a pre-embedded data bus.

10. A self-powered temperature monitoring element, characterized in that, The steps for implementing the method according to any one of claims 1 to 9 include: Thermally conductive encapsulation housing for contact with the interior of concrete and conduction of heat; Thermoelectric conversion module, wherein the thermoelectric conversion module is a thermoelectric generator, the temperature sensing end of the thermoelectric generator is tightly thermally coupled to the thermally conductive encapsulation shell, and the reference end of the thermoelectric generator is provided with a heat insulation structure; An energy management and storage circuit, connected to the output of the thermoelectric conversion module, is used to boost, rectify, and manage the generated electrical energy, and store the processed electrical energy in the energy storage module; the energy management and storage circuit includes a maximum power point tracking circuit, used to dynamically adjust the input impedance to extract the maximum power from the thermoelectric conversion module; The temperature sensing module is connected to the energy management and storage circuit, and collects and processes temperature data after receiving power. A communication unit, connected to the temperature sensing module, is used to send the processed temperature data outward; the communication unit includes a low-power wireless communication module and a reserved wired communication interface, forming a dual-mode redundant communication mechanism.

Images