Concrete condensation state prediction and maintenance method based on multi-sensor fusion

Abstract

The invention relates to the technical field of concrete performance detection, and discloses a concrete condensation state prediction and maintenance method based on multi-sensor fusion. The method aims at solving the problems that an existing concrete condensation state judgment method is destructive and cannot reflect the real state in a structure in real time, and maintenance measures cannot dynamically adapt to actual working condition changes. The method comprises the following steps: deploying a multi-sensor combination in concrete to collect data in real time, pre-processing the data, inputting the pre-processed data into a long-short-term memory network model, and predicting a condensation state and early strength; and the maintenance equipment is automatically controlled according to the prediction result. According to the method, feedback optimization is also performed on the model by utilizing the final actual measurement intensity. Lossless and real-time monitoring and accurate prediction of the internal state of the concrete are realized, automatic and refined closed-loop regulation and control can be performed on the maintenance operation, and the scientificity and efficiency of construction management are improved.

Description

Technical Field

[0001] This invention relates to the field of concrete performance testing technology, and in particular to a method for predicting and curing the setting state of concrete based on multi-sensor fusion. Background Technology

[0002] In modern civil engineering, concrete is the most fundamental and core building material. The process of transforming freshly mixed, plastic concrete into a load-bearing, hard solid—the setting and hardening process—has a decisive impact on the construction progress, structural safety, and long-term durability of the entire project. Currently, construction sites generally rely on traditional methods to determine the setting state of concrete. For example, this involves preparing test blocks according to specifications and periodically testing them with a penetration resistance meter to determine the initial and final setting times, relying on subjective judgment based on the personal experience of construction workers. Furthermore, concrete curing typically employs fixed age standards, such as continuous watering and air drying for 7 to 14 days after final setting. While these long-standing methods have ensured project quality to some extent, they also have inherent limitations.

[0003] The actual setting and hardening process of concrete is a complex physicochemical reaction, and its rate is dynamically affected by various factors such as the type of cement, admixtures, mix proportions, and the ever-changing environmental conditions at the construction site, including temperature and humidity. Traditional fixed testing methods and curing procedures cannot respond to these changes in real time and accurately. Therefore, those skilled in the art have always faced a substantial technical problem that urgently needs to be solved: how to achieve non-destructive, real-time tracking and accurate prediction of the actual performance evolution process of concrete inside the main structure, and based on this, to dynamically and intelligently control curing measures in a closed-loop manner, thereby maximizing construction efficiency and resource utilization while ensuring project quality. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, in order to solve the problems that existing methods for judging the setting state of concrete are destructive, cannot reflect the real internal state of the structure in real time, and that curing measures cannot dynamically adapt to changes in actual working conditions, this invention provides a method for predicting and curing the setting state of concrete based on multi-sensor fusion.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for predicting and curing the setting state of concrete based on multi-sensor fusion, which includes the following steps: S1. Before concrete pouring, a sensor unit including a temperature sensor, a resistivity sensor and an ultrasonic transceiver probe is deployed inside the concrete structure to collect temperature data, resistivity data and ultrasonic propagation data reflecting the concrete hydration process in real time. S2. Preprocess the real-time acquired temperature data, resistivity data, and ultrasonic propagation data. The preprocessing includes noise filtering and data normalization. S3. The preprocessed data is input into a pre-trained long short-term memory network model, which then outputs the prediction results of the setting state and early strength of the concrete in real time. S4. Based on the prediction results, control the curing equipment associated with the concrete structure to perform intelligent curing operations; S5. After the project is completed, core samples of the concrete structure are drilled and compressive strength tests are conducted to determine the actual strength value. The actual strength test value is compared with the prediction result, and the comparison result is used to retrain the long short-term memory network model to optimize the prediction performance.

[0007] As a preferred embodiment of the concrete setting state prediction and curing method based on multi-sensor fusion described in this invention, the following steps are taken: Before concrete pouring, a sensor unit including a temperature sensor, a resistivity sensor, and an ultrasonic transceiver probe is deployed inside the concrete structure, and temperature data, resistivity data, and ultrasonic wave propagation data reflecting the concrete hydration process are collected in real time. The deployment inside the concrete structure specifically involves placing the sensor unit assembly at the center of the thickest part of the concrete structure's cross-section; before pouring the concrete, the sensor unit assembly is tied and fixed to the reinforcing steel, and its wiring connections and cables are waterproofed and sealed; the sensor unit assembly also includes a surface humidity sensor.

[0008] As a preferred embodiment of the concrete setting state prediction and curing method based on multi-sensor fusion described in this invention, the real-time acquired temperature data, resistivity data, and ultrasonic propagation data are preprocessed. The preprocessing includes noise filtering and data normalization, and the specific steps are as follows: The noise filtering uses a smoothing filtering algorithm, and the data normalization uses minimum-maximum normalization processing.

[0009] As a preferred embodiment of the concrete setting state prediction and curing method based on multi-sensor fusion described in this invention, the preprocessed data is input into a pre-trained long short-term memory network model, which then outputs real-time prediction results of the concrete setting state and early strength. The specific steps are as follows: The pre-training includes: synchronously collecting sensor data of concrete test blocks in the laboratory and setting time data measured using a penetration resistance meter, and using the setting time data as a label to conduct supervised learning training on the long short-term memory network model; the prediction results specifically include: a qualitative judgment of the current setting stage, a quantitative prediction of the remaining time to reach the initial and final setting states, and a prediction curve for future strength development.

[0010] As a preferred embodiment of the concrete setting state prediction and curing method based on multi-sensor fusion described in this invention, the method involves controlling the curing equipment associated with the concrete structure to perform intelligent curing operations based on the prediction results. The specific steps are as follows: The curing equipment includes a misting spray device; the intelligent curing operation includes automatically activating the misting spray device according to preset rules when the concrete is predicted to be in a stage of water loss risk and the detection value of the surface humidity sensor is lower than a preset humidity threshold; the curing equipment includes an electric heating insulation device; the intelligent curing operation includes automatically activating the electric heating insulation device to heat and cure the concrete structure according to preset rules when the temperature data is lower than a preset minimum curing temperature.

[0011] As a preferred embodiment of the concrete setting state prediction and curing method based on multi-sensor fusion described in this invention, the method includes: after the project is completed, core samples of the concrete structure are drilled and compressive strength tests are performed to determine the actual strength value. The actual strength test value is then compared with the prediction result, and the comparison result is used to retrain the long short-term memory network model to optimize the prediction performance. The specific steps are as follows: The retraining of the long short-term memory network model is performed using incremental learning, which adds the actual intensity values ​​and their corresponding sensor data as new samples to the original training set.

[0012] The beneficial effects of this invention are: This invention utilizes multiple sensors, including those for temperature, resistivity, and ultrasound, deployed within concrete to collect multi-dimensional physical data in real time. After preprocessing, the data is input into a pre-trained long short-term memory (LSTM) network model to predict the concrete's setting state and early strength in real time. Based on the prediction results, the system automatically controls curing equipment for intelligent curing. After the project is completed, the model is optimized using core drilling measurements, forming a closed-loop adaptive system. This method achieves non-destructive, real-time monitoring of the concrete's internal state, improves the accuracy of setting state and early strength predictions, and enables automated, precise, and dynamic control of curing operations according to actual needs, thereby enhancing the scientific nature of construction management and resource utilization efficiency. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of the overall process of the present invention; Figure 2 This is a flowchart of step S2, data preprocessing, in Example 1. Figure 3 This is a flowchart of steps S3 and S4 of Example 1 for prediction and maintenance control. Figure 4 This is a flowchart of the model lifecycle and optimization process for steps S3 and S5 in Example 1. Detailed Implementation

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0017] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0018] Example 1 Reference Figures 1-4 This is the first embodiment of the present invention, which provides a method for predicting and curing the setting state of concrete based on multi-sensor fusion, including the following steps: S1. Before concrete pouring, a sensor unit including a temperature sensor, resistivity sensor, and ultrasonic transceiver probe is deployed inside the concrete structure to collect temperature data, resistivity data, and ultrasonic wave propagation data reflecting the concrete hydration process in real time. The specific operation steps are as follows: Deployed inside the concrete structure, specifically, the sensor unit assembly is deployed at the center of the thickest part of the concrete structure section; before pouring the concrete, the sensor unit assembly is tied and fixed to the reinforcing steel, and its wiring and cables are waterproofed and sealed; the sensor unit assembly also includes a surface humidity sensor.

[0019] It should be noted that the sensing unit assembly includes a temperature sensor for monitoring the heat of hydration, a resistivity sensor for characterizing changes in pore structure and ion concentration, and a pair of ultrasonic transceiver probes for detecting the development of material mechanical properties. In this embodiment, the sensing unit is deployed at the center of the thickest section of the concrete structure. This is the area where the structural hydration reaction is most intense and the heat release is most concentrated, and its temperature history best represents the adiabatic hydration state of the concrete. At the same time, this location is also the potential risk point with the largest internal and external temperature difference, making monitoring crucial for controlling temperature cracks.

[0020] To ensure the sensors can survive the immense impact of concrete pouring and the highly alkaline corrosive environment during the hardening process, protective measures were taken during deployment: the sensor units were bundled and fixed to the reinforcing bars using non-conductive nylon cable ties to ensure they would not shift during the pouring process; and epoxy resin was used to waterproof and seal the wiring points and cable leads to prevent moisture intrusion that could cause short circuits and signal distortion.

[0021] It should also be noted that, to more comprehensively assess the early cracking risk of the structure, the sensing unit assembly in this embodiment also includes strain sensors. Working in conjunction with temperature sensors, the strain sensors can distinguish between strain caused by temperature changes and strain caused by the chemical shrinkage of the concrete itself, thus providing direct data for monitoring overall shrinkage and providing early warning of cracking risk. All sensors are connected to a field data acquisition terminal via shielded cables. This terminal collects data at a preset frequency (e.g., initially every five minutes) and transmits the data stream to a cloud server in real time via a wireless network, laying the foundation for subsequent analysis.

[0022] S2. Preprocess the real-time acquired temperature data, resistivity data, and ultrasonic propagation data. Preprocessing includes noise filtering and data normalization. The specific steps are as follows: The raw data stream collected from the site contains non-ideal factors such as electromagnetic interference from the construction site and sensor noise, making it unsuitable for direct and accurate modeling and analysis. Therefore, preprocessing is performed before it enters the prediction model. In this embodiment, the preprocessing process includes two core steps: noise filtering and data normalization. Noise filtering employs a smoothing filtering algorithm, and data normalization uses minimum-maximum normalization.

[0023] It should be noted that the system initiates a noise filtering procedure on the received real-time data sequence. This procedure employs a smoothing filtering algorithm to eliminate meaningless high-frequency spikes and minor jitters in the data curve, while preserving the smooth changes that reflect the true trend of the hydration process, making the processed data curve smoother and clearer. Its calculation process can be expressed by the following formula:

[0024] in: : for at time The calculated smoothed data value.

[0025] : This is the window size for the moving average, determining the number of data points used to calculate the average, with a value ranging from 3 to 7.

[0026] : is the summation symbol, indicating that the following terms are summed. Subscript =0 represents the starting term for the summation, with the superscript indicating the term. This represents the term that ends the summation.

[0027] : This is the index variable for summation, starting from 0 and taking values ​​1, 2, ..., until... .

[0028] : The terms to be summed, representing the time from the current moment. Looking back The original measurements for each time step.

[0029] It should also be noted that, to address the issue of differing physical dimensions and vastly different numerical ranges among various sensors (for example, temperature values ​​are typically in the tens, while resistivity values ​​can reach tens of thousands), the system performs data normalization on the filtered data. A minimum-maximum normalization method is used to linearly map all data, regardless of their original numerical values, to a unified interval of [0,1]. The purpose of this step is to eliminate the difference in magnitude between different feature data, preventing the model from developing biases towards certain features during training due to differences in numerical values, and ensuring that information from all dimensions can be learned and utilized by the model. The calculation process can be represented by the following formula:

[0030] in: : The normalized value. is a unitless pure number whose value is limited to the range [0,1].

[0031] This refers to the raw measurement value acquired by the sensor at a specific moment. For a temperature sensor, The temperature range is typically between -10°C and 90°C, therefore It can be set to -10. It can be set to 90; for resistivity sensors, The range is typically in the range of 0.1 Up to 50 Between, therefore It can be set to 0.1. It can be set to 50; for ultrasonic wave speeds, it is typically between 1500 m / s (close to the wave speed in water) and 5000 m / s (after hardening), therefore It can be set to 1500. It can be set to 5000.

[0032] This refers to the theoretical minimum or preset lower limit value that this type of sensor may reach during the entire monitoring period.

[0033] This refers to the theoretical maximum value or preset upper limit that this type of sensor may reach during the entire monitoring period.

[0034] S3. Input the preprocessed data into the pre-trained Long Short-Term Memory (LSTM) network model. The LSM network model will then output the prediction results of the concrete's setting state and early strength in real time. The specific operation steps are as follows: Pre-training includes: synchronously collecting sensor data from concrete specimens in the laboratory and setting time data measured using a penetration resistance meter, and using the setting time data as labels to conduct supervised learning training on the long short-term memory network model; the prediction results specifically include: a qualitative judgment of the current setting stage, a quantitative prediction of the remaining time to reach the initial and final setting states, and a prediction curve for future strength development.

[0035] It should be noted that the Long Short-Term Memory (LSTM) network model, as a recurrent network model, excels in processing and predicting time-series data with long-term dependencies, which aligns well with the physicochemical process of concrete setting and hardening.

[0036] It should also be noted that the model's pre-training was completed in a laboratory environment: multiple sets of concrete specimens with different mix proportions were prepared, and sensor combinations identical to those used in the field were embedded in the specimens to synchronously record sensor data throughout the entire process from mixing to hardening; the penetration resistance of the specimens was measured at regular intervals using a penetration resistance meter to calibrate the true time points of initial and final setting. These sensor data sequences with precise time labels constituted the "standard answer" dataset for model learning. Through supervised learning training, the model was able to master the complex mapping relationship from multi-dimensional sensor data to the setting state.

[0037] The model can instantly output a set of prediction results containing rich decision-making information based on real-time input data. The results are specifically manifested in the following ways: First, a qualitative judgment of the current setting stage, such as "plastic state" or "initial setting period"; Second, a quantitative prediction of the remaining time to reach the initial and final setting states, presented intuitively in the form of a countdown; Third, a dynamically updated future strength development prediction curve, predicting the compressive strength value at any future time (such as 24 hours or 72 hours).

[0038] Among them, the normalized multidimensional data stream obtained after preprocessing is at each sampling time It can be organized into a multidimensional input vector. This vector will be fed into a Long Short-Term Memory (LSTM) network model, which can adjust the input vector based on real-time input. It instantly outputs a set of prediction result vectors containing rich decision-making information. The specific composition of the prediction result vector can be represented as follows:

[0039] : For the model at time 1 The output is a vector of predicted results.

[0040] : To determine the current time for the model The probability value that the initial solidification state has been reached. Its value ranges from [0,1].

[0041] : To determine the current time for the model The probability value of reaching the final solidification state. Its value ranges from [0,1].

[0042] : The model's prediction from the current time The time required to reach the final solidification state, in hours. .

[0043] : This represents the compressive strength value predicted by the model 24 hours later based on the current state, in megapascals (MPa). The value range is 0 Up to 30 Between these values, the value increases over time.

[0044] The formula here is a structured representation of the model's output information. It shows that the model's output at any given time contains a comprehensive set of information, including classification prediction (the probability of reaching a certain state), regression prediction (remaining time), and intensity prediction (future intensity value).

[0045] S4. Based on the prediction results, control the curing equipment associated with the concrete structure to perform intelligent curing operations. The specific operation steps are as follows: The curing equipment includes a misting spray device; the intelligent curing operation includes automatically activating the misting spray device according to preset rules when the concrete is predicted to be in a stage of water loss risk and the detection value of the surface humidity sensor is lower than the preset humidity threshold; the curing equipment includes an electric heating insulation device; the intelligent curing operation includes automatically activating the electric heating insulation device to heat and cure the concrete structure according to preset rules when the temperature data is lower than the preset minimum curing temperature.

[0046] It should be noted that the intelligent maintenance operation in this embodiment is achieved through an embedded, hierarchical maintenance rule base. The rule base can comprehensively judge the model prediction results and real-time environmental data, and resolve possible decision conflicts according to the preset "rule priority", ensuring that the maintenance action most beneficial to the structure is always executed.

[0047] To address the most common early-stage water loss cracking risk in concrete, the sensing unit assembly also includes surface humidity sensors deployed on the structural surface, and a solenoid-valve-controlled atomizing spray system is provided on the construction site. Medium-priority moisturizing curing rules are set in the curing rule library: When the model predicts that the concrete is in a "plastic state" and the surface humidity sensor detects a value lower than the preset 85%RH humidity threshold, the system controller automatically starts high-frequency atomized spraying to sprinkle water.

[0048] When the model predicts that the system has entered the "initial condensation stage," which is most sensitive to moisture, the rule base will raise the humidification standard and start high-frequency spraying when the surface humidity is below 90%RH.

[0049] When the concrete enters the "hardened state" and the model predicts that the strength has not yet reached 70% of the design strength, the moisture retention standard is appropriately relaxed to 70%RH, and low-frequency spraying is used for routine curing.

[0050] To optimize resources, when the model predicts that the intensity has reached or exceeded 70% of the design intensity, low-priority rules will be triggered, and the system will automatically stop or significantly reduce maintenance operations.

[0051] Furthermore, for large-volume concrete structures, controlling the internal and external temperature difference is crucial to preventing temperature cracks. The curing rule library includes a rule based on temperature difference control. The system calculates the internal and external temperature difference in real time using the following formula:

[0052] in: : for time The temperature difference between the inside and outside of the concrete, expressed in degrees Celsius (°C).

[0053] : For temperature sensors deployed at the center of the structure at all times The measured core temperature.

[0054] : For temperature sensors deployed on the surface of the structure at any time The measured surface temperature.

[0055] The maintenance system will continuously monitor The value, once it exceeds the allowable upper limit required by the specification (e.g. When the temperature reaches 25℃, the system will automatically trigger corresponding temperature control measures, such as starting the cooling water pipes embedded in the structure for circulation and cooling, or covering the surface of the structure for insulation, thereby controlling the temperature difference within a safe range.

[0056] It should also be noted that, to address the risks of extreme temperatures, the rule base has established survival protection rules with the highest priority. For example, to address the risk of frost damage during winter construction, electric heating insulation devices are installed on site, and the rule base has set an anti-freezing rule: regardless of the state of the concrete, as long as the core temperature collected by the temperature sensors deployed inside the structure remains below 5°C, the system will activate the electric heating insulation devices to heat it. Similarly, to prevent internal damage caused by excessively high core temperatures, if the core temperature exceeds 65°C, the system will also activate cooling measures such as cooling water circulation with the highest priority. This can be illustrated by the following example in Table 1:

[0057] Table 1 As shown in Table 1, the system comprehensively judges the real-time monitoring and prediction data according to the set priorities. For example, the rules for preventing freezing and preventing core overheating have the highest priority. Under normal temperatures, the system matches different humidity thresholds and spraying strategies according to the different life cycle stages of the concrete (plastic state, initial setting stage, hardened state) and the predicted strength development, thereby achieving refined and intelligent on-demand curing.

[0058] S5. After the project is completed, core samples of the concrete structure are drilled and compressive strength tests are conducted to determine the actual strength value. The actual strength test value is then compared with the predicted result. The comparison result is used to retrain the Long Short-Term Memory Network model to optimize the prediction performance. The specific operation steps are as follows: The retraining of the Long Short-Term Memory network model is carried out by incremental learning, which adds the actual intensity values ​​and their corresponding sensor data as new samples to the original training set.

[0059] It should be noted that after the project is completed and the concrete reaches the specified age (usually 28 days), core samples are drilled from non-critical load-bearing parts of the structure and sent to the laboratory for standard compressive strength testing to determine the final actual strength value.

[0060] This actual strength value, obtained through physical testing and representing the final quality of the project, will be compared with the strength value predicted by the system based on sensor data at the time of the test to evaluate the accuracy of the model in this application. The complete sequence of sensor data recorded throughout the project will be paired with this actual strength result to form a new training sample derived from real-world conditions. The system will use incremental learning to supplement this new sample into the existing model training set, retraining the Long Short-Term Memory (LSTM) network model. By continuously absorbing real-world cases from different projects, materials, and environments, the model can continuously learn, self-correct, and evolve, constantly optimizing its predictive performance and enhancing its adaptability.

[0061] Among them, the accuracy of the model in this application is evaluated, and the commonly used indicator is the mean absolute error (MAE). The calculation process is as follows:

[0062] Mean absolute error, in megapascals (MPA). The smaller the value, the more accurate the model prediction.

[0063] : The total number of core samples used for comparison.

[0064] : is the summation symbol, indicating that all samples are summed.

[0065] : for the first The actual compressive strength value of each core sample obtained through a pressure test.

[0066] : for the corresponding The predicted intensity value output by the model at the specified age.

[0067] Example 2 To further illustrate the present invention, the following example of a construction project for a large-volume concrete bridge pier cap in winter will be used to describe the technical solution of the present invention in more detail, which is the second embodiment of the present invention.

[0068] In this embodiment, a project team for a large bridge project needed to carry out winter construction on the large-volume concrete abutment (design dimensions 15m×10m×3m, C40 concrete) of its No. 3 pier, and ensure that its early performance met design requirements. Construction took place on December 5, 2023, when the nighttime ambient temperature could drop to -5°C.

[0069] S1. Construct a real-time sensing network that penetrates deep into the structure: Prior to the concrete pouring on December 5, the project team built a real-time multi-physics sensing network specifically for this pier structure.

[0070] At the heart of the network is a master sensing unit assembly that integrates a high-precision PT100 temperature sensor, a four-probe resistivity sensor, a pair of piezoelectric ceramic ultrasonic transceivers, and a vibrating wire strain gauge. The master sensing unit is deployed on a steel cage at the geometric center of the foundation (1.5 meters below the bottom surface) to capture key data on the overall hydration process of the structure.

[0071] Considering the temperature difference between inside and outside and the need for surface maintenance, the network also includes two auxiliary measuring points: an auxiliary temperature sensor located at the corner of the top surface of the foundation, and a capacitive surface humidity sensor directly exposed to the top surface of the foundation.

[0072] All sensing units were securely tied to the reinforcing bars with nylon cable ties, and their wiring connections were rigorously sealed with epoxy resin for waterproofing, ensuring survival and stable data transmission during subsequent construction and hardening processes.

[0073] S2. Cleaning and standardizing multi-source heterogeneous data: Since the pouring began at 2 p.m. on December 5, 2023, all sensors have been collecting data at 5-minute intervals, uploading real-time data to the cloud server via on-site wireless data acquisition terminals.

[0074] Upon receiving the raw data stream, the server immediately initiates a preprocessing procedure. For example, at some point, on-site welding work might cause the resistivity sensor to return a reading as high as 1000. The instantaneous pulse value is identified and discarded by the program's outlier removal module.

[0075] Use window size The moving average filtering algorithm with a value of 5 smooths the entire data sequence, generating a smoother curve that better reflects the true hydration trend.

[0076] The system standardizes the filtered data from each stream. For example, it standardizes the core temperature measurement of 65℃ at a given moment through... The calculation of (65-(-10)) / (90-(-10)) converts the data to a normalized value of 0.75. Through this step, all data with different dimensions are converted into dimensionless values ​​in the interval [0,1], forming a standardized multidimensional input vector that can be directly used by the model.

[0077] S3. State prediction and interpretation based on long short-term memory network model: Standardized data streams are continuously fed into a long short-term memory network model that has been pre-trained in the laboratory.

[0078] About 6 hours after the pouring was completed (8 pm on December 5), the predicted result vector output by the model was parsed and presented to the on-site engineer on the project management APP: "Current state: plastic state; core temperature rise rate: 2.5℃ / h; predicted final setting time: about 8 hours and 15 minutes". At 3 a.m. the next day, the APP automatically pushed a warning: "Attention: The structure will enter the final setting stage within 1 hour. Please prepare for surface polishing or roughening." At this time, the remaining final setting time displayed on the interface has been dynamically updated to "55 minutes".

[0079] Meanwhile, the strength development prediction curve on the interface shows that the strength is expected to reach 25MPa at 3 days, which meets the minimum strength requirements for hoisting the upper structure and provides a time basis for the project manager to formulate subsequent construction plans.

[0080] S4. The linkage maintenance equipment realizes closed-loop intelligent maintenance: On the night of December 5th, when the ambient temperature dropped to -2℃, the temperature sensor reading on the surface of the foundation dropped to 4.8℃, which was lower than the preset minimum curing temperature threshold of 5℃. The system automatically triggered the linkage rule and sent an "start" command to the intelligent controller of the electric heating insulation device to start heating curing until the surface temperature rose back to above 8℃ before automatically stopping.

[0081] By the afternoon of the following day, the system had calculated the internal and external temperature difference in real time. The temperature reached 30℃, exceeding the upper limit of 25℃ required by the standard. The system immediately issued the highest level "temperature difference exceeding limit" alarm to the project manager's mobile APP and the on-site audible and visual alarm at the same time, and automatically turned on the circulating water pump of the pre-buried cooling water pipe to remove the excess heat from the core area through water flow, and control the internal and external temperature difference within a safe range.

[0082] On the third day after the pouring, the surface humidity sensor detected that the humidity dropped rapidly to 75%. The system judged that there was a risk of surface dehydration and cracking, and then automatically started the atomizing spray device to maintain the surface of the foundation in an intermittent mode of "spraying for 2 minutes and pausing for 15 minutes".

[0083] S5. Model validation and iterative optimization based on experimental results: Twenty-eight days after the foundation was cured (January 2, 2024), the project team drilled three core samples from non-critical load-bearing parts of the foundation and sent them to the laboratory for compressive strength testing. The average actual strength value was 48.5 MPa.

[0084] The measured values ​​were input into the system backend, and the system compared them with the predicted intensity value (47.0 MPa) of the model at 28 days of age. The mean absolute error (MAE) of this prediction was calculated to be 1.5 MPa, which verifies that the model has high accuracy.

[0085] The complete sensor dataset from the entire process of this bridge pier and foundation project, from S1 to S4, including extreme weather and complex working conditions, along with this 48.5MPa "real-world label," was used as a real-world engineering case study. It was incrementally added to the original training set of the Long Short-Term Memory (LSTM) network model, triggering a background retraining process. After this optimization, its predictions will be more accurate when facing similar large-volume concrete projects in winter.

[0086] In summary, this invention utilizes multiple sensors, including those for temperature, resistivity, and ultrasound, deployed within concrete to collect multi-dimensional physical data in real time. After preprocessing, the data is input into a pre-trained long short-term memory (LSTM) network model to predict the concrete's setting state and early strength in real time. Based on the prediction results, the system automatically controls curing equipment for intelligent curing. After the project is completed, the model is optimized using core drilling measurements, forming a closed-loop adaptive system. This achieves non-destructive, real-time monitoring of the concrete's internal state, improves the accuracy of setting state and early strength predictions, and enables automated, precise, and dynamic control of curing operations according to actual needs, thereby enhancing the scientific nature of construction management and resource utilization efficiency.

[0087] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting and curing the setting state of concrete based on multi-sensor fusion, characterized in that, Includes the following steps: S1. Before concrete pouring, a sensor unit including a temperature sensor, a resistivity sensor and an ultrasonic transceiver probe is deployed inside the concrete structure to collect temperature data, resistivity data and ultrasonic propagation data reflecting the concrete hydration process in real time. S2. Preprocess the real-time acquired temperature data, resistivity data, and ultrasonic propagation data. The preprocessing includes noise filtering and data normalization. S3. The preprocessed data is input into a pre-trained long short-term memory network model, which then outputs the prediction results of the setting state and early strength of the concrete in real time. S4. Based on the prediction results, control the curing equipment associated with the concrete structure to perform intelligent curing operations; S5. After the project is completed, core samples of the concrete structure are drilled and compressive strength tests are conducted to determine the actual strength value. The actual strength test value is compared with the prediction result, and the comparison result is used to retrain the long short-term memory network model to optimize the prediction performance.

2. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The deployment inside the concrete structure specifically means that the sensing units are deployed in combination at the center of the thickest part of the concrete structure cross-section.

3. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 2, characterized in that, The deployment also includes: before pouring concrete, binding and fixing the sensor unit assembly to the reinforcing steel, and waterproofing and sealing its wiring and cables.

4. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The noise filtering uses a smoothing filtering algorithm, and the data normalization uses minimum-maximum normalization processing.

5. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The pre-training includes: synchronously collecting sensor data of concrete test blocks in the laboratory and setting time data measured using a penetration resistance meter, and using the setting time data as a label to conduct supervised learning training on the long short-term memory network model.

6. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The prediction results specifically include: a qualitative assessment of the current solidification stage, a quantitative prediction of the remaining time to reach the initial and final solidification states, and a prediction curve for future intensity development.

7. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The sensing unit assembly also includes a surface humidity sensor; the curing equipment includes a misting spray device; the intelligent curing operation includes automatically activating the misting spray device according to preset rules when it is predicted that the concrete is in a stage of water loss risk and the detection value of the surface humidity sensor is lower than a preset humidity threshold.

8. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The curing equipment includes an electric heating insulation device; the intelligent curing operation includes automatically activating the electric heating insulation device to heat and cure the concrete structure when the temperature data is lower than the preset minimum curing temperature, according to preset rules.

9. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The retraining of the long short-term memory network model is performed using incremental learning, which adds the actual intensity values ​​and their corresponding sensor data as new samples to the original training set.

10. The method for predicting and curing the setting state of concrete based on multi-sensor fusion as described in claim 1, characterized in that, The sensing unit assembly also includes a strain sensor for real-time acquisition of strain data of the concrete to monitor its shrinkage and cracking risks.

Images