A method for detecting the quality of construction concrete

Abstract

The application relates to the technical field of concrete quality detection, and discloses a method for detecting the quality of concrete used in building construction, which comprises the following steps: arranging a multi-modal sensing network to collect thermodynamic, acoustic and electrical full-time domain data in real time; performing spatiotemporal alignment and characteristic processing on multi-source heterogeneous data to generate a normalized multi-dimensional time sequence characteristic vector sequence; inputting the sequence into a physical information neural network prediction model to solve a three-dimensional compressive strength field and an internal stress field of concrete in real time; and generating a regulation and control instruction based on the solving result to drive an automatic curing system to realize closed-loop adaptive control. Through the deep learning fusion of multi-modal perception and physical mechanism constraint, the method realizes micro-crack risk feedforward early warning and precise curing, and improves the durability of a concrete structure and the construction quality.

Description

Technical Field

[0001] This invention belongs to the field of concrete quality testing technology, and specifically relates to a method for testing the quality of concrete in building construction. Background Technology

[0002] With the increasing demands for structural safety and durability in modern construction engineering, concrete, as a core building material, requires quality control throughout the entire construction process. The development of concrete strength is dominated by the hydration reaction process, which is accompanied by significant exothermic effects and volumetric strain. These two processes are highly coupled in the early stages (from initial setting to near final setting), jointly determining the formation of internal microstructures and the evolution of defects. Traditional quality testing methods mainly rely on the compressive strength testing of specimens under standard curing conditions or subsequent non-destructive testing methods such as rebound and ultrasonic testing. These methods cannot obtain real-time hydration dynamics information under the actual construction site conditions, resulting in strength estimation lagging significantly behind the actual construction progress.

[0003] Coordinated monitoring of the early hydration heat release rate and internal strain field of concrete has become a key path for accurately assessing strength development. Existing sensing technologies mostly employ thermocouples or resistance strain gauges for single-point temperature or strain measurements, which suffer from drawbacks such as susceptibility to electromagnetic interference, poor durability, and inability to be distributed and deployed, making it difficult to construct a continuous sensing network covering key areas of the component. More importantly, current detection systems lack the ability to deeply integrate temperature and strain field data to invert hydration kinetic parameters. This results in curing strategies (such as heat preservation, moisture retention, and demolding time) still relying on experience or fixed schedules, unable to be dynamically adjusted according to the actual hydration process of the concrete.

[0004] Current technologies cannot achieve high-precision, long-distance, and interference-resistant synchronous sensing of both temperature and strain parameters, nor have they established a real-time mapping model between hydration heat release characteristics and strength growth. This results in a lack of timely warnings in the early stages of temperature difference stress or shrinkage strain accumulation, easily inducing irreversible microcracks and weakening the overall structural integrity and service life. This problem is particularly prominent during construction of large-volume concrete or under extreme climatic conditions, urgently requiring a quality inspection method that can be embedded in the structure itself, capture the evolution of hydration multiphysics fields in real time, and drive intelligent maintenance decisions. Summary of the Invention

[0005] The technical problem to be solved by this invention is that existing concrete quality testing technologies cannot establish a real-time and accurate intrinsic correlation model between the physical characterization of the early hydration process and the development of later mechanical properties, resulting in a significant lag in concrete curing decisions and an inability to proactively control the risk of microcracks caused by concentrated hydration heat and internal and external temperature differences.

[0006] To address the aforementioned technical problems, this invention provides a method for inspecting the quality of concrete in building construction. This method utilizes a multimodal, distributed sensor network deployed within the concrete structure to collect real-time, synchronous, full-time-domain thermodynamic, acoustic, and electrical evolution data of the concrete from the moment of pouring, constructing a three-dimensional spatiotemporal data field. Furthermore, this method establishes a hybrid prediction model integrating physical mechanism constraints and data-driven approaches. This model takes the multimodal sensor data stream as input and, through deep temporal feature extraction and regularization constraints of the physical control equations, calculates in real-time the compressive strength development curve and internal stress distribution cloud map at any location within the concrete. Finally, based on the real-time calculated strength and stress state, this method generates and outputs precise control commands for an automated curing system, achieving closed-loop, adaptive regulation of the concrete curing process, thereby fundamentally eliminating curing decision lag and suppressing the generation of early microcracks.

[0007] According to an aspect of the present invention, a method for testing the quality of concrete in building construction is provided, comprising the following steps:

[0008] A multimodal sensor network is deployed on the steel reinforcement skeleton inside the concrete structure to be poured, and multi-source heterogeneous sensor data covering the entire curing cycle is acquired.

[0009] The multi-source heterogeneous sensing data is subjected to spatiotemporal reference alignment and feature processing to generate a normalized multi-dimensional temporal feature vector sequence.

[0010] The multidimensional temporal feature vector sequence is input into a pre-trained physical information neural network prediction model to calculate the three-dimensional compressive strength field and three-dimensional internal stress field of the concrete structure in real time.

[0011] Based on the real-time calculation results of the three-dimensional compressive strength field and the three-dimensional internal stress field, a set of control commands for controlling the automated maintenance system is generated and output.

[0012] As one embodiment of the present invention, the acquisition of multi-source heterogeneous sensing data specifically includes:

[0013] The temperature and strain time series data of each measuring point along the fiber optic path inside the concrete structure are continuously measured at a preset spatial sampling interval using an embedded distributed fiber optic grating sensing array, forming a thermodynamic data stream. The distributed fiber optic grating sensing array is deployed in three dimensions along the key stress path of the structure to capture the temperature gradient and strain gradient inside the structure.

[0014] Multiple pairs of piezoelectric ceramic transducers are arranged in predetermined positions inside the structure to periodically emit and receive ultrasonic pulses, measure the propagation time of ultrasonic waves in the concrete medium, and calculate the time sequence data of ultrasonic wave velocity to form an acoustic data stream. The arrangement positions of the piezoelectric ceramic transducer pairs cover the core load-bearing area and the geometric change zone of the structure.

[0015] An embedded four-probe resistivity sensor array is used to measure the time-series data of volume resistivity in multiple representative areas inside concrete using a constant AC excitation method, forming an electrical data stream. The four-probe resistivity sensor array is made of corrosion-resistant metal electrodes to ensure long-term measurement stability in a strongly alkaline environment.

[0016] By setting up environmental monitoring units at the construction site, real-time data on air temperature, relative humidity, and wind speed of the construction environment are collected and used as external boundary conditions for model calculations.

[0017] Furthermore, the step of performing spatiotemporal reference alignment and feature processing on the multi-source heterogeneous sensing data to generate a normalized multidimensional temporal feature vector sequence specifically includes:

[0018] All data streams collected by the sensors are synchronized and aligned according to a unified global timestamp, and the data at different sampling rates are resampled using a cubic spline interpolation algorithm to form an original dataset with a unified time resolution.

[0019] The discrete temperature data acquired by the distributed fiber optic grating sensing array is processed using the Kriging space interpolation algorithm to construct a continuous, time-evolving three-dimensional temperature field function. Key thermodynamic features are extracted from the three-dimensional temperature field function, including: the peak temperature of the adiabatic temperature rise curve, the time to reach the peak temperature, the heating rate, the cooling rate, and the maximum temperature stress gradient calculated from the temperature field gradient.

[0020] The ultrasonic wave velocity time series data is subjected to first-order difference processing to obtain the wave velocity change rate characteristics; the volume resistivity time series data is subjected to logarithmic transformation and first-order difference processing to obtain the resistivity logarithmic change rate characteristics.

[0021] The thermodynamic characteristics, wave velocity change rate characteristics, resistivity logarithmic change rate characteristics, and environmental monitoring data are combined into a single, high-dimensional feature vector at each time step, and the vector is then subjected to max-min normalization to finally form the multi-dimensional time-series feature vector sequence.

[0022] In a preferred embodiment of the present invention, the physical information neural network prediction model is a deep hybrid network structure, and its construction and training process includes:

[0023] A recurrent neural network encoder with a gated recurrent unit as its core is constructed to receive the multidimensional temporal feature vector sequence, learn its deep temporal dependence and nonlinear dynamic evolution law, and output a hidden state vector containing historical information.

[0024] A decoder consisting of fully connected layers is constructed. The decoder receives the hidden state vector as input and outputs preliminary prediction results of the three-dimensional compressive strength field and three-dimensional internal stress field of concrete at the current moment.

[0025] A composite loss function is defined for training the model. The composite loss function is composed of a weighted average of a data-driven loss term and a physical mechanism constraint loss term. The data-driven loss term is defined by calculating the root mean square error between the model's preliminary prediction results and the measured compressive strength data of laboratory-cured test blocks. The physical mechanism constraint loss term is defined by substituting the model's preliminary prediction results into the residuals generated by the coupled partial differential equations of cement hydration kinetics and thermodynamics. The equations specifically include the Arrhenius hydration rate equation, Fourier's law of heat conduction, and thermoelastic constitutive relations.

[0026] By employing the backpropagation algorithm and the adaptive moment estimation optimizer, the composite loss function is minimized iteratively, and the network weight parameters inside the recurrent neural network encoder and decoder are adjusted until the model converges, thus obtaining the final physical information neural network prediction model.

[0027] Furthermore, based on the real-time calculation results of the three-dimensional compressive strength field and the three-dimensional internal stress field, a set of control commands for controlling the automated maintenance system is generated and output, specifically including:

[0028] During the real-time solution process of the model, the maximum tensile stress value in the three-dimensional internal stress field is continuously monitored, and the theoretical value of the tensile strength of the concrete at the current moment is calculated based on the three-dimensional compressive strength field.

[0029] A decision logic module is established. When the maximum tensile stress value exceeds the preset safety factor threshold of the current theoretical tensile strength value, it is determined that there is a risk of microcracks and a maintenance intervention command is triggered.

[0030] The maintenance intervention command is transmitted to the automated maintenance execution system, which performs precise operations according to the command content. When a risk of micro-cracks is determined, if the risk is caused by excessive internal temperature difference, the command activates the cooling water circulation system deployed on the formwork surface to reduce the surface temperature and decrease the internal and external temperature difference. If the risk is caused by excessive surface water loss, the command activates the intelligent spraying system covering the concrete surface to increase surface humidity and replenish moisture.

[0031] The control command is a set of pulse width modulation signals, the duty cycle of which is proportional to the degree to which the maximum tensile stress value exceeds the safety threshold, thereby realizing proportional control of the intensity of maintenance measures and forming a complete closed-loop feedback control loop.

[0032] According to another aspect of the present invention, a concrete quality testing system for building construction is also provided, comprising:

[0033] A multimodal sensor network acquisition module is configured to be deployed inside the concrete structure to be poured and to collect multi-source heterogeneous sensor data covering the entire curing cycle in real time. The data includes thermodynamic, acoustic and electrical evolution data.

[0034] The data processing and feature engineering module, which is electrically connected to the multimodal sensor network acquisition module, is configured to perform spatiotemporal reference alignment and feature processing on the multi-source heterogeneous sensor data to generate a normalized multidimensional time-series feature vector sequence.

[0035] The model prediction and solution module is electrically connected to the data processing and feature engineering module. It has a pre-trained physical information neural network prediction model embedded in it and is configured to receive the multi-dimensional time-series feature vector sequence and calculate the three-dimensional compressive strength field and three-dimensional internal stress field of the concrete structure in real time.

[0036] The maintenance decision and control module, which is electrically connected to the model prediction and solution module, is configured to generate and output a set of control commands for controlling the automated maintenance system based on the real-time solution results of the three-dimensional compressive strength field and the three-dimensional internal stress field.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] 1. By fusing multimodal sensor data, this invention constructs a panoramic, multi-dimensional physical profile of the early hydration process of concrete. The completeness and accuracy of the information acquisition far exceed that of existing technologies that rely solely on temperature measurement.

[0039] 2. This invention creatively introduces a physical information neural network model, which couples the data-driven deep learning paradigm with mature first principles of physics, ensuring that the strength and stress prediction results not only fit the data but are also self-consistent in terms of physical mechanisms, greatly improving the model's generalization ability and prediction accuracy.

[0040] 3. This invention realizes full-process automation and closed-loop control from data acquisition and state calculation to maintenance decision-making, transforming the traditional concrete maintenance management mode, which relies on manual experience, is lagging and discrete, into a real-time data-driven, feedforward, continuous and precise control mode.

[0041] 4. By performing real-time calculations of the internal stress field and providing early warnings of micro-crack risks, this invention can guide automated curing systems to intervene at the optimal time, precisely control the temperature and humidity differences between the inside and outside of the concrete, and fundamentally inhibit the formation of early shrinkage cracks, thus ensuring the long-term durability and safety of the building structure. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the overall technical solution architecture of a method for testing the quality of concrete in building construction proposed in this invention.

[0043] Figure 2 This is a schematic diagram of the core principle framework of the hybrid prediction model that integrates physical mechanism constraints and data-driven approaches in this invention.

[0044] Figure 3 This is a logical flowchart of the multimodal sensor network deployment and multi-source heterogeneous sensor data acquisition in this invention.

[0045] Figure 4 This is a flowchart illustrating the logical flow of spatiotemporal reference alignment and feature processing of multi-source heterogeneous sensor data in this invention. Detailed Implementation

[0046] Please refer to Figure 1-4 This invention provides a method for inspecting the quality of concrete in building construction. Its core lies in the real-time acquisition of thermodynamic, acoustic, and electrical evolution data of concrete from the moment it is poured using a multimodal sensor network. Based on a physical information neural network model, it performs real-time calculations of the three-dimensional compressive strength field and three-dimensional internal stress field within the concrete, ultimately generating automated curing control commands to achieve feedforward proactive control of micro-crack risks. This method strictly follows the sequence of steps.

[0047] S1. A multimodal sensing network is deployed on the internal steel reinforcement skeleton of the concrete structure to be poured, acquiring multi-source heterogeneous sensing data covering the entire curing cycle. The multimodal sensing network consists of a distributed fiber optic grating sensor array, paired piezoelectric ceramic transducers, a four-probe resistivity sensor array, and an environmental monitoring unit. The distributed fiber optic grating sensor array is deployed in three-dimensional space along the critical stress path of the concrete structure, with a spatial sampling interval set at 0.5 meters to ensure accurate capture of the spatial distribution characteristics of temperature and strain gradients within the structure.

[0048] The array is fixed to the surface of the reinforcing steel frame before concrete pouring and protected by a special sleeve to prevent mechanical damage during the pouring process. Piezoelectric ceramic transducers are deployed in transmitter-receiver pairs in the core load-bearing area and areas of geometric abrupt change in the structure. The linear distance between each pair of transducers does not exceed 3 meters to ensure effective penetration of the ultrasonic signal in the concrete medium and a high signal-to-noise ratio. The four-probe resistivity sensor array consists of four corrosion-resistant metal electrodes arranged in a straight line at equal intervals, with a spacing of 10 centimeters. The entire array is encapsulated in a waterproof and insulating housing and deployed in representative areas within the concrete, such as beam-column joints, the center of the slab thickness, and edge areas.

[0049] The environmental monitoring unit is installed near the concrete structure at the construction site. It is dustproof and waterproof and continuously collects air temperature, relative humidity, and wind speed data as external boundary conditions required for subsequent model calculations. All sensing units are connected to the central processing unit through industrial-grade data acquisition equipment. The data acquisition frequency is uniformly set to once per minute to ensure the integrity and synchronization of data across the entire time domain.

[0050] S2, the multi-source heterogeneous sensor data is aligned to a spatiotemporal reference and characterized to generate a normalized multi-dimensional time-series feature vector sequence. First, the data streams collected by all sensors are synchronized and aligned according to a unified global timestamp, with the completion time of concrete pouring as zero and using Coordinated Universal Time (UTC). Since the original sampling rates of different sensors differ—for example, the distributed fiber grating array samples once per second, while the environmental monitoring unit samples once per minute—a cubic spline interpolation algorithm is used to upsample low-frequency data and downsample high-frequency data, ultimately forming an original dataset with a unified time resolution of one minute per time step. Second, for the discrete temperature measurement point data output by the distributed fiber grating sensor array, a Kriging spatial interpolation algorithm is used for three-dimensional reconstruction, constructing a continuous, time-evolving three-dimensional temperature field function. .

[0051] Key thermodynamic characteristic parameters were extracted from this temperature field function, including the peak temperature of the adiabatic temperature rise curve. This value represents the highest temperature recorded at all locations within the entire structure during the curing period; the time it takes to reach the peak temperature. Defined as from the moment of pouring until The time interval between the first occurrences; the rate of heating , calculated as Difference from the initial temperature divided by Cooling rate , calculated as The temperature difference after 24 hours divided by the corresponding time interval; maximum temperature stress gradient. This is obtained by calculating the spatial gradient of the temperature field function and taking the maximum value of its magnitude. Next, the ultrasonic propagation time series measured by the piezoelectric ceramic transducer is processed, first converted into ultrasonic wave velocity. Then, perform a first-order difference operation on it to obtain the wave velocity change rate characteristics. Volume resistivity measured by a four-probe resistivity sensor The sequence is obtained by first performing a natural logarithmic transformation. Then perform a first-order difference to obtain the logarithmic rate of change of resistivity characteristic. Finally, at each time step... The above-extracted thermodynamic characteristics Acoustic characteristics Electrical characteristics And environmental monitoring data {air temperature, relative humidity, wind speed} are combined into a single high-dimensional feature vector. The vector has nine dimensions. It is then subjected to min-max normalization, which linearly maps the value of each dimension to the interval 0-1. The formula is as follows:

[0052] ;

[0053] in and These are the minimum and maximum values ​​of this dimension across the entire training dataset, respectively. After this processing, a normalized multidimensional temporal feature vector sequence is formed. .

[0054] S3, the multidimensional temporal feature vector sequence is input into a pre-trained physical information neural network prediction model to calculate the three-dimensional compressive strength field and three-dimensional internal stress field of the concrete structure in real time. The physical information neural network prediction model is a deep hybrid network structure, and its construction process is as follows:

[0055] First, a recurrent neural network encoder with gated recurrent units at its core is constructed. This encoder consists of two stacked layers of gated recurrent units, each with 128 hidden units. It receives normalized multidimensional temporal feature vector sequences and learns their deep temporal dependencies and nonlinear dynamic evolution patterns. At a time step... The encoder outputs a hidden state vector containing historical information. .

[0056] Secondly, a decoder consisting of three fully connected layers is constructed, with 256, 512, and 1024 neurons in each layer, and a modified linear unit (MRU) activation function. The decoder receives the hidden state vector. As input, and output the three-dimensional compressive strength field of concrete at the current moment. With three-dimensional internal stress field Preliminary forecast results.

[0057] The prediction results are represented in the form of a voxel grid with a grid resolution of 0.5 meters, consistent with the spatial sampling interval of the distributed fiber Bragg grating array.

[0058] The model training process uses a composite loss function for optimization. Data-driven loss terms loss term constrained by physical mechanism Weighted composition, i.e. ,in and The preset weighting coefficients are 0.7 and 0.3 respectively. Data-driven loss term. It is defined as the root mean square error between the model-predicted compressive strength value and the measured compressive strength value of the laboratory-cured specimen at the corresponding age.

[0059] Physical mechanism constraint loss term The residuals are defined by substituting the predicted temperature, strength, and stress fields into a set of pre-defined physical control equations. These physical control equations include: the Arrhenius hydration rate equation, which describes the change in cement hydration degree with temperature and time; Fourier's law of heat conduction, which constrains the spatial evolution of the temperature field to satisfy heat flow conservation; and thermoelastic constitutive relations, which ensure that the predicted stress, strain, and temperature fields satisfy the thermoelastic coupling relationship of the material.

[0060] During training, the backpropagation algorithm and adaptive moment estimation optimizer are used to iteratively update the network weights. The initial learning rate is set to 0.001, and the batch size is 32. The model is considered converged when the loss function on the validation set no longer decreases for 10 consecutive rounds, thus obtaining the final physical information neural network prediction model. In practical applications, this model is deployed on an edge computing device in the field, receiving the feature vector sequence output from step S2 and outputting the solution results of the three-dimensional compressive strength field and the three-dimensional internal stress field in real time at a frequency of once per minute.

[0061] S4. Based on the real-time calculation results of the three-dimensional compressive strength field and the three-dimensional internal stress field, a set of control commands for the automated maintenance system is generated and output. This step first involves continuously monitoring the maximum tensile stress value in the three-dimensional internal stress field during the real-time model calculation. Meanwhile, based on the three-dimensional compressive strength field Using empirical formulas Calculate the theoretical value of the tensile strength of the concrete at the current moment. ,in The value is a material constant, which is 0.33 for ordinary silicate concrete. A decision logic module is established, which has a built-in preset safety factor threshold γ, with a value of 0.8.

[0062] When the maximum tensile stress value is monitored in real time Exceed When the system detects a risk of microcracks inside the concrete structure, it immediately triggers a maintenance intervention command.

[0063] The maintenance intervention command is transmitted to the automated maintenance execution system, which includes a cooling water circulation system deployed on the formwork surface and an intelligent spraying system covering the concrete surface. The decision logic module further analyzes the causes of microcrack risk: if the three-dimensional temperature field shows a significant temperature gradient inside the structure, and the area of ​​maximum tensile stress coincides with the high-temperature core area, the risk is determined to be caused by excessive internal temperature difference, and the generated command is to start the cooling water circulation system; if environmental monitoring data shows that the relative humidity of the air is too low and the wind speed is too high, and the resistivity of the concrete surface area rises sharply, the risk is determined to be caused by excessive surface water loss, and the generated command is to start the intelligent spraying system.

[0064] The specific form of the control command is a set of pulse width modulation signals with a duty cycle of It is directly proportional to the degree to which the maximum tensile stress exceeds the safety threshold, and the calculation formula is:

[0065] ;

[0066] in The preset sensitivity coefficient is set to 0.2. This pulse width modulation signal directly drives the power controller of the cooling water pump or spray solenoid valve, thereby achieving proportional control of the intensity of the curing measures. The entire process constitutes a complete closed-loop feedback control circuit, ensuring that the timing and intensity of the curing intervention are precisely matched with the actual stress state inside the concrete.

[0067] As one embodiment of the present invention, the physical mechanism constraint loss term of the physical information neural network prediction model The specific calculations involve the following three core equations. The first equation is the Arrhenius hydration rate equation, which is expressed as follows:

[0068] ;

[0069] in, For hydration degree, Pre-exponential factor, For activation energy, The gas constant is... Absolute temperature This represents the reaction order. The model predicts the temperature field. With intensity field The dynamic relationship described by this equation must be satisfied, and the residual is... The first component.

[0070] The second equation is Fourier's law of heat conduction, and its differential form is:

[0071] ;

[0072] in, For concrete density, For specific heat capacity, Thermal conductivity, This is the hydration heat source term. The temperature field predicted by the model must satisfy this partial differential equation, and its numerical residuals on the discrete grid constitute... The second component.

[0073] The third equation is the thermoelastic constitutive relation, and its simplified form is:

[0074] ;

[0075] in, For stress tensor, Let be the elastic stiffness tensor. For strain tensor, The coefficient of thermal expansion is For reference temperature, This is the Kronecker notation. The stress field, strain field (obtainable from distributed fiber Bragg grating strain data), and temperature field predicted by the model must satisfy this relationship, and their residuals constitute... The third component. By summing the squared residuals of these three terms with weights, the complete physical mechanism constraint loss term is obtained. .

[0076] In one embodiment of the present invention, the multimodal sensor network acquisition module, data processing and feature engineering module, model prediction and solution module, and maintenance decision and control module together constitute a concrete quality inspection system for building construction. The multimodal sensor network acquisition module is responsible for the acquisition and initial transmission of all sensor data in step S1. The data processing and feature engineering module performs all data processing tasks in step S2, including time synchronization, spatial interpolation, feature extraction, and normalization. The model prediction and solution module has a trained physical information neural network prediction model embedded within it, dedicated to performing real-time solution in step S3.

[0077] The maintenance decision and control module implements the risk assessment and instruction generation logic in step S4, and interfaces with the on-site cooling water circulation system and intelligent sprinkler system via industrial communication protocols. As a whole, the system realizes an end-to-end automated process from the perception of raw physical quantities to the output of advanced decisions, enabling full-process monitoring and maintenance control of concrete quality without human intervention.

[0078] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for testing the quality of concrete in building construction, characterized in that, include: A multimodal sensor network is deployed on the steel reinforcement skeleton inside the concrete structure to be poured, and multi-source heterogeneous sensor data covering the entire curing cycle is acquired. The multi-source heterogeneous sensor data includes thermodynamic data stream, acoustic data stream, electrical data stream and environmental monitoring data. The multi-source heterogeneous sensing data is subjected to spatiotemporal reference alignment and feature processing to generate a normalized multi-dimensional temporal feature vector sequence. The multidimensional temporal feature vector sequence is input into a pre-trained physical information neural network prediction model to calculate the three-dimensional compressive strength field and three-dimensional internal stress field of the concrete structure in real time. Based on the real-time calculation results of the three-dimensional compressive strength field and the three-dimensional internal stress field, a set of control commands for controlling the automated maintenance system is generated and output.

2. The method for testing the quality of concrete in building construction according to claim 1, characterized in that, Acquiring the multi-source heterogeneous sensing data includes: The temperature and strain time series data of each measuring point along the fiber optic path inside the concrete structure are continuously measured at a preset spatial sampling interval using an embedded distributed fiber optic grating sensing array, forming the thermodynamic data stream; the distributed fiber optic grating sensing array is deployed in three dimensions along the key stress path of the structure. Multiple pairs of piezoelectric ceramic transducers are arranged in predetermined positions inside the structure to periodically emit and receive ultrasonic pulses, measure the propagation time of ultrasonic waves in the concrete medium, and calculate the time sequence data of ultrasonic wave velocity to form the acoustic data stream; the arrangement positions of the piezoelectric ceramic transducer pairs cover the core load-bearing area and the geometric change zone of the structure. The electrical data stream is formed by measuring the volume resistivity time series data of multiple representative regions inside concrete using an embedded four-probe resistivity sensor array with constant AC excitation. The four-probe resistivity sensor array is composed of corrosion-resistant metal electrodes. By setting up an environmental monitoring unit at the construction site, real-time data on air temperature, relative humidity, and wind speed in the construction environment are collected as environmental monitoring data.

3. The method for testing the quality of concrete in building construction according to claim 2, characterized in that, The multi-source heterogeneous sensing data is subjected to spatiotemporal reference alignment and feature processing to generate a normalized multidimensional temporal feature vector sequence, including: All data streams collected by the sensors are synchronized and aligned according to a unified global timestamp, and the data at different sampling rates are resampled using a cubic spline interpolation algorithm to form an original dataset with a unified time resolution. The discrete temperature data collected by the distributed fiber optic grating sensing array is processed by the Kriging space interpolation algorithm to construct a continuous, time-evolving three-dimensional temperature field function. Key thermodynamic features are extracted from the three-dimensional temperature field function, including the peak temperature of the adiabatic temperature rise curve, the time to reach the peak temperature, the heating rate, the cooling rate, and the maximum temperature stress gradient calculated from the temperature field gradient. The ultrasonic wave velocity time series data is subjected to first-order difference processing to obtain the wave velocity change rate characteristics; the volume resistivity time series data is subjected to logarithmic transformation and first-order difference processing to obtain the resistivity logarithmic change rate characteristics. The thermodynamic features, wave velocity change rate features, resistivity logarithmic change rate features, and environmental monitoring data are combined at each time step to form a high-dimensional feature vector, and the vector is then subjected to max-min normalization to form the multi-dimensional time-series feature vector sequence.

4. The method for testing the quality of concrete in building construction according to claim 3, characterized in that, The physical information neural network prediction model is a deep hybrid network structure, and its construction and training process includes: A recurrent neural network encoder with a gated recurrent unit as its core is constructed to receive the multidimensional temporal feature vector sequence, learn its deep temporal dependence and nonlinear dynamic evolution law, and output a hidden state vector containing historical information. A decoder consisting of fully connected layers is constructed. The decoder receives the hidden state vector as input and outputs preliminary prediction results of the three-dimensional compressive strength field and three-dimensional internal stress field of concrete at the current moment. A composite loss function is defined for training the model, which is composed of a weighted average of a data-driven loss term and a physical mechanism constraint loss term.

5. The method for testing the quality of concrete in building construction according to claim 4, characterized in that, The data-driven loss term is defined by calculating the root mean square error between the preliminary prediction results of the model and the measured compressive strength data of the laboratory synchronous curing test blocks; the physical mechanism constraint loss term is defined by substituting the preliminary prediction results of the model into the residuals generated by substituting them into the cement hydration kinetic equations and the thermodynamic coupled partial differential equations, which include the Arrhenius hydration rate equation, Fourier's heat conduction law, and thermoelastic constitutive relations.

6. The method for testing the quality of concrete in building construction according to claim 5, characterized in that, By employing the backpropagation algorithm and the adaptive moment estimation optimizer, the composite loss function is minimized iteratively, and the network weight parameters inside the recurrent neural network encoder and decoder are adjusted until the model converges, thus obtaining the final physical information neural network prediction model.

7. The method for testing the quality of concrete in building construction according to claim 6, characterized in that, The physical mechanism constraint loss term is obtained by substituting the model prediction results into the Arrhenius hydration rate equation, Fourier's law of heat conduction, and the thermoelastic constitutive relation, and then calculating the weighted sum of the squared residuals. The Arrhenius hydration rate equation is used to describe the change of hydration degree with temperature and time. The Fourier law of heat conduction is used to constrain the spatial evolution of the temperature field to satisfy the conservation of heat flow. The thermoelastic constitutive relation is used to ensure that the predicted stress field, strain field, and temperature field satisfy the thermoelastic coupling relationship of the material.

8. The method for testing the quality of concrete in building construction according to claim 7, characterized in that, The recurrent neural network encoder contains two stacked gated recurrent units, with 128 hidden units in each layer; the decoder consists of three fully connected layers, with 256, 512 and 1024 neurons in each layer, and the activation function is a modified linear unit; the model outputs a three-dimensional compressive strength field and a three-dimensional internal stress field represented in voxel grid form, with a grid resolution of 0.5 meters.

9. The method for testing the quality of concrete in building construction according to claim 8, characterized in that, Based on the real-time calculation results of the three-dimensional compressive strength field and the three-dimensional internal stress field, a set of control commands for controlling the automated maintenance system is generated and output, including: During the real-time solution process of the model, the maximum tensile stress value in the three-dimensional internal stress field is continuously monitored, and the theoretical value of the tensile strength of the concrete at the current moment is calculated based on the three-dimensional compressive strength field. A decision logic module is established. When the maximum tensile stress value exceeds the preset safety factor threshold of the current theoretical tensile strength value, it is determined that there is a risk of microcracks and a maintenance intervention command is triggered. The maintenance intervention command is transmitted to the automated maintenance execution system, which performs precise operations according to the command content. When a risk of micro-cracks is determined, if the risk is caused by excessive internal temperature difference, the command activates the cooling water circulation system deployed on the formwork surface; if the risk is caused by excessive surface water loss, the command activates the intelligent spraying system covering the concrete surface.

10. The method for testing the quality of concrete in building construction according to claim 9, characterized in that, The control command is a set of pulse width modulation signals, the duty cycle of which is proportional to the degree to which the maximum tensile stress value exceeds the safety threshold, thereby realizing proportional control of the intensity of maintenance measures and forming a complete closed-loop feedback control loop.

Images