Method for detecting compressive strength of building concrete member based on multi-technology fusion

Abstract

The invention discloses a method for detecting compressive strength of a building concrete member based on multi-technology fusion, and belongs to the technical field of concrete members. The invention discloses a building concrete member compressive strength detection method based on multi-technology fusion. The method comprises the following steps: collecting multi-dimensional detection data; obtaining a feature vector; fusion features are obtained; obtaining a preliminary compressive strength result; optimizing deep learning feature fusion model parameters; a concrete member digital twinborn model is constructed, and strength dynamic prediction and tracing are achieved; a final detection result is obtained; and displaying through a visual terminal. The method solves the problem that the detection precision of the high-strength and special-shaped concrete member is difficult to guarantee in the prior art, effectively eliminates the influence of interference factors such as environment temperature, member water content and detection angle on the detection result, greatly improves the credibility of the detection result, can comprehensively show the evolution process of the member strength, and improves the detection accuracy. Manual operation errors are greatly reduced, and the detection efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of concrete component technology, specifically to a method for testing the compressive strength of building concrete components based on the integration of multiple technologies. Background Technology

[0002] The compressive strength of concrete components is a core indicator for assessing the safety, durability, and reliability of building structures, directly affecting the quality and service life of construction projects. In scenarios such as construction quality acceptance, health monitoring of existing building structures, and repair and reinforcement of old buildings, the testing of the compressive strength of concrete components is an indispensable and crucial step.

[0003] Chinese Patent Publication No. CN113702223B discloses a method and system for testing the compressive strength of concrete components based on the rebound method. The method includes measuring the high-strength rebound value and the medium-strength rebound value at rebound test points in multiple rebound test zones on the concrete component; calculating the high-strength average rebound value and the medium-strength average rebound value, as well as the high-strength rebound distribution aggregation index and the medium-strength rebound distribution aggregation index, based on the high-strength average rebound value and the medium-strength average rebound value of all rebound test points in the target test area; optimizing the high-strength average rebound value and the medium-strength average rebound value and calculating the compressive strength value of the concrete component.

[0004] The aforementioned patents rely solely on rebound data in practical applications, without establishing a dynamic environmental error compensation mechanism. The test results are easily affected by factors such as temperature, humidity, and surface flatness of the components, making it difficult to guarantee the testing accuracy of high-strength and irregularly shaped concrete components. Therefore, they do not meet the current needs. In response, we propose a method for testing the compressive strength of building concrete components based on the integration of multiple technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a method for testing the compressive strength of building concrete components based on the integration of multiple technologies. This method can comprehensively capture the multifaceted characteristics of concrete components, effectively eliminate the influence of interference factors such as ambient temperature, component moisture content, and testing angle on the test results, and significantly improve the reliability of the test results. By updating model parameters in real time, it realizes dynamic prediction and traceability of the strength of concrete components throughout their entire life cycle from the construction stage to the service stage, and can comprehensively show the evolution process of component strength. By adaptively generating the optimal distribution path of testing points, it ensures that testing points cover the entire area and key defect areas are tested in a focused manner, which greatly reduces human operation errors, improves testing efficiency, and solves the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for testing the compressive strength of building concrete components based on the integration of multiple technologies, comprising the following steps: S1: Perform full-area scanning of building concrete components, simultaneously collect multi-dimensional detection data, and perform data cleaning and standardization processing on the collected multi-dimensional detection data; S2: Extract deep features of each detection data through a convolutional neural network to obtain feature vectors. The feature vectors include: ultrasonic feature vector, rebound feature vector, electromagnetic induction feature vector and infrared thermal imaging feature vector. S3: Construct a deep learning feature fusion model, input the extracted feature vectors into the deep learning feature fusion model for weighted fusion and cross-validation, and obtain the fused features; S4: Use edge computing to preprocess and fuse feature vectors and upload them to the cloud. The cloud matches the benchmark database and performs dynamic calibration in combination with environmental parameters to obtain preliminary compressive strength results. S5: Obtain the true strength value of the concrete component, use the true strength value to adjust the weight coefficients of the deep learning feature fusion model, and optimize the parameters of the deep learning feature fusion model. S6: Integrate multi-dimensional detection data, calibration data, and true strength values ​​to construct a digital twin model of concrete components, enabling dynamic strength prediction and traceability; S7: Identify the shape and defects of concrete components, generate the optimal detection path using a genetic algorithm based on the identification results, and obtain the final detection result by combining environmental data for secondary compensation. S8: The final compressive strength test results, test point distribution, component defect location, and full life cycle strength evolution curve are displayed through a visualization terminal, and a test report is generated at the same time.

[0007] Preferably, S3 specifically includes: A deep learning model based on the attention mechanism is constructed, and the variances of the extracted ultrasonic feature vector, rebound feature vector, electromagnetic induction feature vector and infrared thermal imaging feature vector are normalized to obtain the initial weight coefficients. Collect historical test data and corresponding strength values ​​of different types of components. After standardizing the historical test data, associate it with the strength values ​​to form a training dataset. Dynamic weight calibration based on attention mechanism inputs the constructed training dataset into the deep learning model; The information contribution of each feature vector in the detection of different types of components is analyzed by deep learning model, and an error evaluation matrix is ​​constructed by combining historical detection data. The initial weight coefficients are calibrated based on the error evaluation matrix. The calibrated weight coefficients are then multiplied element-wise with the corresponding feature vectors to obtain the weighted feature data of each feature vector. The weighted feature data are summed to generate a preliminary fused feature vector. The validity of the preliminary fused feature vector is then checked until a valid fused feature vector is obtained.

[0008] Preferably, the calculation process for the weight coefficients of the attention mechanism is as follows: The variance of each eigenvector is normalized to obtain the initial weights; An error assessment matrix is ​​constructed by combining historical detection data. The initial weights are then corrected based on the error assessment matrix to obtain the final weight coefficients. The dataset is divided into training and validation sets using k-fold cross-validation, and the effectiveness of the fused feature vectors is judged by the error rate of the validation set.

[0009] Preferably, S4 specifically includes: An edge computing module and cloud big data collaborative architecture are constructed. The edge computing module performs real-time preprocessing of the fused feature vector, and the preprocessed fused feature vector and the collected raw detection data are uploaded to the cloud. A concrete strength benchmark database is built in the cloud. A dynamic matching algorithm is used to match the fused feature vector uploaded from the site with the benchmark feature vector in the benchmark database to obtain the corresponding benchmark strength value and environmental impact coefficient. By combining the environmental parameters uploaded by the edge computing module, dynamic calibration is performed to obtain the preliminary compressive strength test results after calibration.

[0010] Preferably, the edge computing module uses an industrial-grade edge server, and the cloud-based big data architecture uses a distributed storage system to store the concrete strength benchmark database. The construction process of the benchmark database is as follows: Collect concrete strength test data from different regions and different engineering types, and after normalizing the collected data, establish a mapping relationship between the benchmark feature vector and the benchmark strength value; The cosine similarity algorithm is used to calculate the cosine similarity between the field fusion feature vector and the benchmark feature vector. The benchmark data with the highest similarity is selected for weighted averaging to obtain the matching benchmark strength value. The on-site environmental parameters are compared with the environmental impact coefficients in the benchmark database, and adjustments are made based on the comparison results.

[0011] Preferably, S6 specifically includes: Collect life-cycle correlation data of concrete components, construct digital twin models of components using the life-cycle correlation data, and update the parameters of digital twin models at a preset frequency; Real-time environmental dynamic data of the area where the component is located is collected, and the real-time collected environmental dynamic data is input into the digital twin model of the component to output the strength prediction value of the component at different future time periods. By linking the detection data, strength values, model parameters, and environmental data at each time point during the construction phase and in the relevant area, a full life-cycle data chain is constructed.

[0012] Preferably, S7 specifically includes: Multi-angle and full-coverage image acquisition is performed on concrete components, and three-dimensional coordinate data of each point on the surface of the component are collected simultaneously to form the original image set and three-dimensional point cloud data of the concrete component. The original image set was preprocessed to extract the actual outline of the concrete component, and the spatial position information of each area on the surface of the concrete component was marked by combining the three-dimensional point cloud data. Surface defects are identified in the preprocessed images, and the location, size, shape information and corresponding three-dimensional coordinates of cracks, holes and peeling areas are marked to generate a defect distribution map with spatial location. Based on the actual outline, spatial location information and defect distribution map of the concrete component, the surface of the component is divided into several grid units according to the three-dimensional coordinate range; Set planning constraints for detection points, construct an objective function, use a genetic algorithm to solve for the optimal solution, and output the three-dimensional coordinates of the detection points and the detection order; Real-time acquisition of on-site environmental interference parameters, and calculation of the surface flatness of concrete components based on captured images of concrete components; An environmental interference error database is constructed based on massive historical data. Compensation coefficients are obtained by matching real-time environmental parameters and correcting the preliminary results to obtain the final compressive strength value.

[0013] Preferably, the process of obtaining the final compressive strength value specifically includes: Acquire massive amounts of historical detection data and use this data to construct a database of the correspondence between environmental interference errors; The real-time collected on-site environmental interference parameters are input into the database for matching to obtain the corresponding error compensation coefficients; The preliminary compressive strength test results are retrieved and secondary compensation corrections are performed to eliminate the influence of environmental factors and component surface condition on the test results, thus obtaining the final compressive strength test results of the concrete component.

[0014] Preferably, the intensity dynamic prediction and tracing of S6 specifically includes: Based on acoustic emission data, the acoustic emission event density per unit time is calculated, whereby the event density is the number of acoustic emission events detected within a unit time window; the average event energy, average amplitude, and proportion of high-frequency components are statistically analyzed to characterize the activity level of internal cracks in the component; based on micro-vibration detection data, the rate of change of the natural frequency of micro-vibration, the rate of change of damping ratio, and the change in mode shape correlation coefficient are calculated within adjacent monitoring periods. The acoustic emission event density of the excitation device is jointly mapped with the rate of change of the natural frequency of micro-vibration to form an internal damage index. The internal damage index is expressed as: Internal damage index = α × normalized value of acoustic emission event density + β × relative rate of change of natural frequency + γ × rate of change of damping ratio, where α, β, and γ are weighting coefficients determined based on historical test data or engineering calibration. The internal damage index is used to uniformly quantify damage-sensitive information from different sources into a single time series, constructing an internal damage time series curve. In the initial stage, the current compressive strength value of the component is obtained through a non-destructive testing fusion model. In non-critical parts or during the test stage, some true strength values ​​are obtained as calibration benchmarks. Compressive strength prediction is implemented. Historical strength data, internal damage indicators and environmental impact factors are used as input variables. The predicted strength values ​​at multiple future times are output. The prediction results form a compressive strength evolution curve with time as the horizontal axis and compressive strength as the vertical axis. Correlation analysis is performed on the internal damage time series curve and the compressive strength evolution curve. By calculating the correlation coefficient, lag correlation and trend consistency between the two, the influence law of internal damage changes on strength decay is identified. When the system detects an abnormal increase in the internal damage index in a short period of time, and its growth trend is highly correlated with the downward trend of the strength prediction curve, it is judged as a strength deterioration risk signal, and an early warning of component strength deterioration risk is issued.

[0015] Preferably, the deep learning feature fusion model constructed by S3 adopts a two-layer network structure of a "base model" and a "matching transfer layer", wherein, The base model is responsible for learning the general feature-intensity mapping relationship, and the proportioning transfer layer dynamically corrects the output of the base model according to the input proportioning parameters to achieve adaptive adjustment under different material systems. The mix transfer layer is responsible for automatically entering the online fine-tuning mode when the system detects that the concrete components of the project use a new mix ratio not covered by the database. It uses a small number of samples to quickly train the transfer layer, enabling the deep learning feature fusion model to converge quickly under the new mix ratio conditions. By embedding a state evolution sub-model into the deep learning feature fusion model, the time information of concrete from the initial setting period, early strength development period to long-term service period is transformed into learnable parameters and used for the learning and training of the state evolution sub-model, enabling the state evolution sub-model to dynamically adjust the prediction logic as the concrete age changes.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention can comprehensively capture the multifaceted characteristics of concrete components, effectively eliminating the influence of interference factors such as ambient temperature, component moisture content, and detection angle on the test results, thus greatly improving the reliability of the test results. By updating model parameters in real time, it realizes dynamic prediction and traceability of the strength of concrete components throughout their entire life cycle from the construction stage to the service stage, and can comprehensively show the evolution process of component strength. By adaptively generating the optimal distribution path of detection points, it ensures that the detection points cover the entire area and key defect areas are detected in a focused manner, which greatly reduces human operation errors and improves detection efficiency. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method for testing the compressive strength of building concrete components based on the integration of multiple technologies according to the present invention; Figure 2 This is a flowchart illustrating the specific process of obtaining the fusion feature in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To address the issue that existing testing techniques are susceptible to interference from factors such as temperature, humidity, and the surface smoothness of components, thus compromising the accuracy of testing high-strength, irregularly shaped concrete components, please refer to [link to relevant documentation]. Figures 1-2 This embodiment provides the following technical solution: A method for testing the compressive strength of building concrete components based on the integration of multiple technologies includes the following steps: S1: Perform a full-area scan of the building's concrete components, simultaneously collecting multi-dimensional detection data. The collected multi-dimensional detection data undergoes data cleaning and standardization. This multi-dimensional data includes ultrasonic testing data, rebound testing data, electromagnetic induction testing data, and infrared thermal imaging data. Ultrasonic testing data includes ultrasonic wave propagation speed and amplitude attenuation coefficient; rebound testing data includes rebound value and rebound angle; electromagnetic induction testing data includes magnetic field strength variation and rebar distribution density; and infrared thermal imaging data includes component surface temperature distribution and temperature gradient. S2: Deep features of each detection data are extracted through a convolutional neural network to obtain feature vectors. The feature vectors include: ultrasonic feature vector, rebound feature vector, electromagnetic induction feature vector, and infrared thermal imaging feature vector. Data cleaning adopts the 3σ criterion to remove abnormal data. Standardization adopts the min-max standardization method to map data of different dimensions to the [0,1] interval. S3: Construct a deep learning feature fusion model, input the extracted feature vectors into the deep learning feature fusion model for weighted fusion and cross-validation, and obtain the fused features; S4: Use edge computing to preprocess and fuse feature vectors and upload them to the cloud. The cloud matches the benchmark database and performs dynamic calibration in combination with environmental parameters to obtain preliminary compressive strength results. S5: Take micro-damage samples from non-critical parts of concrete components, conduct compressive strength tests on the core samples, obtain the true value of concrete strength, use the true value of strength as a benchmark, perform error analysis on the preliminary compressive strength test results, adjust the weight coefficients of the deep learning feature fusion model, and optimize the parameters of the deep learning feature fusion model. S6: Integrate multi-dimensional detection data, calibration data, and true strength values ​​to construct a digital twin model of concrete components, enabling dynamic strength prediction and traceability; S7: Identify component shape and defects, generate the optimal detection path using a genetic algorithm based on the identification results, and obtain the final detection result by combining environmental data for secondary compensation. S8: The final compressive strength test results, test point distribution, component defect location, and full life cycle strength evolution curve are displayed through a visualization terminal. At the same time, a test report is generated, which includes basic component information, test methods, test data, test results, accuracy assessment, durability assessment suggestions, and maintenance and reinforcement decision suggestions.

[0020] S3 specifically includes: A deep learning model based on the attention mechanism is constructed, and the variances of the extracted ultrasonic feature vector, rebound feature vector, electromagnetic induction feature vector and infrared thermal imaging feature vector are normalized to obtain the initial weight coefficients. Collect historical test data and corresponding strength values ​​of different types of components. After standardizing the historical test data, associate it with the strength values ​​to form a training dataset. Dynamic weight calibration based on attention mechanism inputs the constructed training dataset into the deep learning model; The information contribution of each feature vector in the detection of different types of components is analyzed by deep learning model, and an error evaluation matrix is ​​constructed by combining historical detection data. The initial weight coefficients are calibrated based on the error evaluation matrix. Feature vectors with smaller detection errors are assigned larger weight coefficients. After calibration, the sum of the weight coefficients of all feature vectors is ensured to be 1. The calibrated weight coefficients are multiplied element-wise with the corresponding feature vectors to obtain the weighted feature data of each feature vector. The weighted feature data are summed to generate a preliminary fused feature vector. The validity of the preliminary fused feature vector is then checked until a valid fused feature vector is obtained.

[0021] The calculation process for the weight coefficients of the attention mechanism is as follows: The variance of each eigenvector is normalized to obtain the initial weights; An error evaluation matrix is ​​constructed by combining historical detection data. The error evaluation matrix reflects the detection error of different feature vectors under different component types. The initial weights are corrected according to the error evaluation matrix to obtain the final weight coefficients. The weight coefficients satisfy the condition that the sum of the weights of each feature vector is 1. The dataset is divided into training and validation sets using k-fold cross-validation. The effectiveness of the fused feature vectors is judged by the error rate of the validation set. When the error rate is greater than 5%, the corresponding feature vectors are removed and the fusion is performed again.

[0022] S4 specifically includes: An edge computing module and cloud big data collaborative architecture are constructed. The edge computing module performs real-time preprocessing of the fused feature vector, and the preprocessed fused feature vector and the collected raw detection data are uploaded to the cloud. A concrete strength benchmark database is built in the cloud. The benchmark database contains concrete strength benchmark data for different mix proportions, different ages, and different curing environments. The concrete strength benchmark data includes benchmark strength values, benchmark feature vectors, and environmental impact coefficients. The cloud uses a dynamic matching algorithm to match the fused feature vector uploaded on-site with the benchmark feature vector in the benchmark database to obtain the corresponding benchmark intensity value and environmental impact coefficient. Dynamic calibration is performed by combining environmental parameters (temperature and humidity) uploaded by the on-site edge computing module to obtain preliminary compressive strength test results after calibration.

[0023] The edge computing module uses industrial-grade edge servers, featuring data caching, real-time preprocessing, and local storage. Preprocessing includes data compression and format conversion. Data compression employs the LZ77 algorithm, achieving a compression ratio of ≥4:1. The cloud-based big data architecture utilizes a distributed storage system to store the concrete strength benchmark database. The construction process of the benchmark database is as follows: Collect concrete strength test data from different regions and different engineering types, and after normalizing the collected data, establish a mapping relationship between the benchmark feature vector and the benchmark strength value; The cosine similarity algorithm is used to calculate the cosine similarity between the field fusion feature vector and the benchmark feature vector. The top 3 sets of benchmark data with the highest similarity are selected and weighted averaged to obtain the matching benchmark strength value. The on-site environmental parameters are compared with the environmental impact coefficients in the benchmark database, and adjustments are made based on the comparison results.

[0024] S6 specifically includes: Collect lifecycle-related data of concrete components, including test data, calibration results, true strength values, as well as component design parameters (concrete mix proportion, component size, reinforcement layout, design strength grade), construction process data (pouring time, vibration parameters, pouring temperature, formwork removal time), curing records (curing temperature, curing humidity, curing duration, curing method), and basic environmental data of the area (climate type of the area where the component is located, historical temperature and humidity statistics). Use the lifecycle-related data to integrate BIM technology and Unity3D engine to build a digital twin model of the component. Update the parameters of the digital twin model at a preset frequency: once every 24 hours during the construction phase and once every 7 days during the area phase. Real-time update is triggered when a strength mutation (mutation amplitude > 10%) is detected. Real-time environmental dynamic data (real-time temperature, humidity, solar radiation intensity, load changes) of the area where the component is located are collected. The real-time collected environmental dynamic data is input into the digital twin model of the component, and the predicted strength values ​​of the component at different future time periods are output. By linking the test data, strength values, model parameters and environmental data at each time point of the construction phase and the phase in which the component is located, a full life cycle data chain is constructed. This allows engineers to query the strength status, data source and environmental background of any stage of the component by time dimension, and realize the complete traceability of the strength evolution process.

[0025] S7 specifically includes: Multi-angle and full-coverage image acquisition is performed on concrete components, and three-dimensional coordinate data of each point on the surface of the component are acquired simultaneously to form the original image set and three-dimensional point cloud data of the concrete component. The original image set was preprocessed to extract the actual outline of the concrete component, and the spatial position information of each area on the surface of the concrete component was marked by combining the three-dimensional point cloud data. Surface defects are identified in the preprocessed images, and the location, size, shape information and corresponding three-dimensional coordinates of cracks, holes and peeling areas are marked to generate a defect distribution map with spatial location. Based on the actual outline, spatial location information and defect distribution map of the concrete component, the surface of the component is divided into several grid units according to the three-dimensional coordinate range; Set the planning constraints for detection points (detection points cover the entire grid, the number of detection points in grid cells of key defect areas is doubled, and the total length of the detection path is minimized), construct the objective function, use a genetic algorithm to solve for the optimal solution, and output the three-dimensional coordinates of the detection points and the detection order; Three to five sensor nodes are evenly distributed in the detection area to collect environmental interference parameters in real time, including temperature and humidity data. At the same time, the surface flatness of the concrete component is calculated based on the captured images of the concrete component (expressed as the maximum deviation value between each point on the surface and the reference plane). The acquisition frequency of all environmental interference parameters is consistent with the detection frequency. An environmental interference error database is constructed based on massive historical data. Compensation coefficients are obtained by matching real-time environmental parameters and correcting the preliminary results to obtain the final compressive strength value.

[0026] To obtain the final compressive strength value, the specific steps include: Acquire massive amounts of historical testing data and use this data to construct a database of environmental interference error correspondences. This database contains the mapping relationship between different combinations of temperature, humidity, and surface smoothness parameters and the corresponding testing errors. The real-time collected on-site environmental interference parameters are input into the database for matching to obtain the corresponding error compensation coefficients; The preliminary compressive strength test results are retrieved and corrected through secondary compensation. The compensation formula is: final strength = calibration strength × (1 + compensation coefficient). This eliminates the influence of environmental factors and component surface condition on the test results, thus obtaining the final compressive strength test results of the concrete component.

[0027] S1 further includes: By arranging multiple high-sensitivity sensors, which have a wide frequency response capability and preferably a frequency range of 20 kHz to 500 kHz, the high-frequency elastic wave signals generated by concrete components under environmental loads, temperature and humidity changes and / or micro-vibration disturbances are captured in real time. The high-frequency elastic wave signals include event count, vibration amplitude, energy, duration and spectral distribution, wherein the spectral distribution is obtained by fast Fourier transform. A micro-vibration excitation device is arranged near the component. The excitation device is preferably an electromagnetic or piezoelectric exciter. A periodic excitation with an amplitude less than the component's own amplitude safety threshold is applied, and the excitation frequency range covers the first few natural frequency ranges of the component. The dynamic response signal of the component under excitation is collected by an accelerometer or displacement sensor. The natural frequency, damping ratio and mode shape variation characteristics of the component are obtained by time-domain-frequency domain analysis. The natural frequency reflects the overall stiffness state of the component, the damping ratio reflects the energy dissipation capacity, and the mode shape variation reflects local damage or stiffness distribution changes. Acoustic emission signals and micro-vibration response signals are stored synchronously according to a unified timestamp, forming a multi-source time-series monitoring dataset.

[0028] The intensity dynamic prediction and tracing of S6 specifically includes: Based on acoustic emission data, the acoustic emission event density per unit time is calculated, whereby the event density is the number of acoustic emission events detected within a unit time window; the average event energy, average amplitude, and proportion of high-frequency components are statistically analyzed to characterize the activity level of internal cracks in the component; based on micro-vibration detection data, the rate of change of the natural frequency of micro-vibration, the rate of change of damping ratio, and the change in mode shape correlation coefficient are calculated within adjacent monitoring periods. The acoustic emission event density of the excitation device is jointly mapped with the rate of change of the natural frequency of micro-vibration to form an internal damage index. The internal damage index is expressed as: Internal damage index = α × normalized value of acoustic emission event density + β × relative rate of change of natural frequency + γ × rate of change of damping ratio, where α, β, and γ are weighting coefficients determined based on historical test data or engineering calibration. The internal damage index is used to uniformly quantify damage-sensitive information from different sources into a single time series, constructing an internal damage time series curve. In the initial stage, the current compressive strength value of the component is obtained through a non-destructive testing fusion model. In non-critical parts or during the test stage, some true strength values ​​are obtained as calibration benchmarks. A time series regression model or a state-space model can be used to predict the compressive strength. Historical strength data, internal damage indicators and environmental impact factors are used as input variables, and the predicted strength values ​​at multiple future times are output. The prediction results form a compressive strength evolution curve with time as the horizontal axis and compressive strength as the vertical axis. Correlation analysis is performed on the internal damage time series curve and the compressive strength evolution curve. By calculating the correlation coefficient, lag correlation and trend consistency between the two, the influence law of internal damage changes on strength decay is identified. When the system detects an abnormal increase in the internal damage index in a short period of time, and its growth trend is highly correlated with the downward trend of the strength prediction curve, it is judged as a strength deterioration risk signal, and an early warning of component strength deterioration risk is issued.

[0029] This solution can issue early warning information in tiers based on preset thresholds. For example, a Level 1 warning is triggered when the growth rate of internal damage indicators exceeds twice the historical average, or when there is an accelerating growth trend for several consecutive periods; a Level 2 warning is triggered when the predicted intensity may fall below the design safety threshold in the future. Warning information is pushed to relevant personnel through terminal interfaces, mobile devices, or management platforms, along with intensity evolution curves, damage time-series curves, and prediction results, to assist in formulating reinforcement, repair, or load-limiting measures.

[0030] Through the above technical solutions, a complete technical chain has been realized, from multi-source sensing and perception, internal damage quantification, dynamic strength prediction, historical status tracing to risk warning. This upgrades the compressive strength of concrete components from traditional "static detection" to "dynamic perception, continuous prediction and traceable management", significantly improving the scientific nature and foresight of structural safety monitoring.

[0031] This solution, building upon the existing fusion of ultrasonic, rebound, electromagnetic, and infrared thermal imaging technologies, introduces acoustic emission detection and micro-vibration response detection technologies to construct a collaborative sensing mechanism for "internal micro-damage—macro-strength." Acoustic emission technology, through the deployment of multiple high-sensitivity sensors, captures in real-time high-frequency elastic wave signals generated by concrete components under environmental loads, temperature and humidity changes, or micro-vibration disturbances. These high-frequency elastic wave signals can include parameters such as event count, amplitude, energy, duration, and spectral distribution. Micro-vibration detection applies low-amplitude periodic excitation to the component using an excitation device, acquiring its natural frequency, damping ratio, and mode shape variation characteristics.

[0032] Introducing an "internal damage evolution sub-model" into the digital twin model, the density of acoustic emission events and the change of micro-vibration frequency are mapped to internal damage indicators to form an internal damage time series curve. This curve is then correlated with the compressive strength evolution curve. When the internal damage indicator shows an abnormal increase, an early warning of the risk of strength deterioration can be given, thus realizing the ability to reverse reasoning from damage to predict strength.

[0033] In the detection process, when an abnormal acoustic emission signal is identified, the detection point planning strategy can be automatically adjusted so that the genetic algorithm prioritizes the densification of detection points in the abnormal area, thereby increasing the detection density of key parts. This solution solves the problem of insufficient perception of "non-penetrating cracks and internal micro-damage" in traditional methods, thus enabling the early detection of potential structural strength defects and having superior promotion and practical value.

[0034] The deep learning feature fusion model constructed by S3 adopts a two-layer network structure of "base model" and "matching transfer layer", wherein, The base model is responsible for learning the general feature-intensity mapping relationship, and the proportioning transfer layer dynamically corrects the output of the base model according to the input proportioning parameters to achieve adaptive adjustment under different material systems. The mix transfer layer is responsible for automatically entering the online fine-tuning mode when the system detects that the concrete components of the project use a new mix ratio not covered by the database. It uses a small number of samples to quickly train the transfer layer, enabling the deep learning feature fusion model to converge quickly under the new mix ratio conditions. By embedding a state evolution sub-model into the deep learning feature fusion model, the time information of concrete from the initial setting period, early strength development period to long-term service period is transformed into learnable parameters and used for the learning and training of the state evolution sub-model, enabling the state evolution sub-model to dynamically adjust the prediction logic as the concrete age changes.

[0035] The state evolution sub-model is defined with state transition equations, which are expressed by the following formula: in, This indicates the strength state of the component at time t; This represents the coefficient of natural variation of a component's strength over time, for example, taking... b represents the predicted strength state of the component; b represents the environmental influence weight, for example, taking... Indicates environmental impact factors, for example, taking Representing state covariance, for example Represents the variance of measurement noise, for example This represents the strength value of the component measured at time t.

[0036] If the previous moment Estimated intensity Based on the estimated intensity from the previous moment, the prediction formula is as follows: and process noise values The predicted intensity state can be calculated. The intensity value obtained by detection Then the calculation yields This means that the strength state of the component at time t can be updated to 43.761 MPa.

[0037] To address the issue of insufficient model generalization ability caused by significant differences in concrete mix proportions and rapid age changes in concrete used in engineering projects, a "multi-age-multi-mix proportion self-learning transfer mechanism" was introduced. This mechanism can store not only baseline strength data in a cloud database but also incorporate mix proportion parameter vectors, including key indicators such as water-cement ratio, sand ratio, admixture type, and mineral admixture proportion, forming joint samples with the detection feature vectors.

[0038] In terms of model structure, a two-layer network structure of "base model + proportion transfer layer" is constructed. The base model is responsible for learning the general feature-intensity mapping relationship, while the proportion transfer layer dynamically corrects the output of the base model based on the input proportion parameters, realizing adaptive adjustment under different material systems. When a new project is detected to use a new proportion not covered in the database, it automatically enters the online fine-tuning mode, using a small number of samples to quickly train the transfer layer, enabling the model to converge quickly under the new proportion conditions.

[0039] In terms of age dimension, an age embedding vector is introduced to transform the time information of concrete from the initial setting stage and early strength development stage to the long-term service life into learnable parameters, enabling the model to dynamically adjust its prediction logic as the age changes. For example, the weight of temperature and humidity features is increased in the early age stage, and the weight of micro-damage and environmental erosion features is increased in the service life stage.

[0040] This solution transforms the detection system from a static model application into a self-evolving engineering system with continuous learning capabilities. It solves the problem of traditional detection models being difficult to transfer between different projects, significantly improving the system's engineering adaptability and long-term usability. Furthermore, this solution introduces a state-space model to describe the evolution of strength over time, and provides corresponding state transition equations to describe this evolution. This enables a quantitative description of the component's strength evolution over time, allowing for state tracking and prediction during component storage and use. If risks are predicted, early warnings can be issued.

[0041] By collecting multi-dimensional data on mechanical response, internal defects, and surface features, and using a feature fusion model to achieve weighted fusion and cross-validation of the data, this method can comprehensively capture the multifaceted characteristics of concrete components. It effectively solves the problem of low detection accuracy for high-strength concrete, irregularly shaped components, and old components. By introducing an edge computing and cloud-based big data collaborative architecture, the detection data can be uploaded to a cloud-based concrete strength benchmark database in real time. A dynamic matching algorithm provides accurate benchmark references for on-site detection data, and real-time calibration is performed in conjunction with on-site environmental parameters. This effectively eliminates the influence of interference factors such as ambient temperature, component moisture content, and detection angle on the detection results, achieving dynamic correction of the detection results and improving their stability. The non-destructive testing fusion model is trained and optimized using the true strength value as a benchmark, achieving single-point true value calibration. The synergy between precision and full-domain non-destructive testing significantly enhances the reliability of test results. By linking test data with component design parameters, construction process data, maintenance records, and service environment data, a digital twin model of the concrete component is constructed. Through real-time updates of model parameters, dynamic prediction and traceability of the concrete component's strength throughout its entire lifecycle, from construction to service, are achieved. This comprehensively demonstrates the evolution of component strength, providing full-cycle data support for component durability assessment. Simultaneously, it can accurately predict the remaining service life of the component. By identifying the component's geometry and surface defect locations, it adaptively generates the optimal distribution path of test points, ensuring full coverage of test points and focused testing of key defect areas. Through secondary correction of test results, no manual intervention is required, significantly reducing human error and improving testing efficiency.

[0042] 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.

[0043] 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.

Claims

1. A method for testing the compressive strength of building concrete components based on the integration of multiple technologies, characterized in that, Includes the following steps: S1: Perform full-area scanning of building concrete components, simultaneously collect multi-dimensional detection data, and perform data cleaning and standardization processing on the collected multi-dimensional detection data; S2: Extract deep features of each detection data through a convolutional neural network to obtain feature vectors. The feature vectors include: ultrasonic feature vector, rebound feature vector, electromagnetic induction feature vector and infrared thermal imaging feature vector. S3: Construct a deep learning feature fusion model, input the extracted feature vectors into the deep learning feature fusion model for weighted fusion and cross-validation, and obtain the fused features; S4: Use edge computing to preprocess and fuse feature vectors and upload them to the cloud. The cloud matches the benchmark database and performs dynamic calibration in combination with environmental parameters to obtain preliminary compressive strength results. S5: Obtain the true strength value of the concrete component, use the true strength value to adjust the weight coefficients of the deep learning feature fusion model, and optimize the parameters of the deep learning feature fusion model. S6: Integrate multi-dimensional detection data, calibration data, and true strength values ​​to construct a digital twin model of concrete components, enabling dynamic strength prediction and traceability; S7: Identify the shape and defects of concrete components, generate the optimal detection path using a genetic algorithm based on the identification results, and obtain the final detection result by combining environmental data for secondary compensation. S8: The final compressive strength test results, test point distribution, component defect location, and full life cycle strength evolution curve are displayed through a visualization terminal, and a test report is generated at the same time.

2. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 1, characterized in that, S3 specifically includes: A deep learning model based on the attention mechanism is constructed, and the variances of the extracted ultrasonic feature vector, rebound feature vector, electromagnetic induction feature vector and infrared thermal imaging feature vector are normalized to obtain the initial weight coefficients. Collect historical test data and corresponding strength values ​​of different types of components. After standardizing the historical test data, associate it with the strength values ​​to form a training dataset. Dynamic weight calibration based on attention mechanism inputs the constructed training dataset into the deep learning model; The information contribution of each feature vector in the detection of different types of components is analyzed by deep learning model, and an error evaluation matrix is ​​constructed by combining historical detection data. The initial weight coefficients are calibrated based on the error evaluation matrix. The calibrated weight coefficients are then multiplied element-wise with the corresponding feature vectors to obtain the weighted feature data of each feature vector. The weighted feature data are summed to generate a preliminary fused feature vector. The validity of the preliminary fused feature vector is then checked until a valid fused feature vector is obtained.

3. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 2, characterized in that, The calculation process for the weight coefficients of the attention mechanism is as follows: The variance of each eigenvector is normalized to obtain the initial weights; An error assessment matrix is ​​constructed by combining historical detection data. The initial weights are then corrected based on the error assessment matrix to obtain the final weight coefficients. The dataset is divided into training and validation sets using k-fold cross-validation, and the effectiveness of the fused feature vectors is judged by the error rate of the validation set.

4. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 1, characterized in that, S4 specifically includes: An edge computing module and cloud big data collaborative architecture are constructed. The edge computing module performs real-time preprocessing of the fused feature vector, and the preprocessed fused feature vector and the collected raw detection data are uploaded to the cloud. A concrete strength benchmark database is built in the cloud. A dynamic matching algorithm is used to match the fused feature vector uploaded from the site with the benchmark feature vector in the benchmark database to obtain the corresponding benchmark strength value and environmental impact coefficient. By combining the environmental parameters uploaded by the edge computing module, dynamic calibration is performed to obtain the preliminary compressive strength test results after calibration.

5. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 4, characterized in that, The edge computing module uses an industrial-grade edge server, and the cloud-based big data architecture uses a distributed storage system to store the concrete strength benchmark database. The construction process of the benchmark database is as follows: Collect concrete strength test data from different regions and different engineering types, and after normalizing the collected data, establish a mapping relationship between the benchmark feature vector and the benchmark strength value; The cosine similarity algorithm is used to calculate the cosine similarity between the field fusion feature vector and the benchmark feature vector. The benchmark data with the highest similarity is selected for weighted averaging to obtain the matching benchmark strength value. The on-site environmental parameters are compared with the environmental impact coefficients in the benchmark database, and adjustments are made based on the comparison results.

6. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 1, characterized in that, S6 specifically includes: Collect life-cycle correlation data of concrete components, construct digital twin models of components using the life-cycle correlation data, and update the parameters of digital twin models at a preset frequency; Real-time environmental dynamic data of the area where the component is located is collected, and the real-time collected environmental dynamic data is input into the digital twin model of the component to output the strength prediction value of the component at different future time periods. By linking the detection data, strength values, model parameters, and environmental data at each time point during the construction phase and in the relevant area, a full life-cycle data chain is constructed.

7. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 1, characterized in that, S7 specifically includes: Multi-angle and full-coverage image acquisition is performed on concrete components, and three-dimensional coordinate data of each point on the surface of the component are collected simultaneously to form the original image set and three-dimensional point cloud data of the concrete component. The original image set was preprocessed to extract the actual outline of the concrete component, and the spatial position information of each area on the surface of the concrete component was marked by combining the three-dimensional point cloud data. Surface defects are identified in the preprocessed images, and the location, size, shape information and corresponding three-dimensional coordinates of cracks, holes and peeling areas are marked to generate a defect distribution map with spatial location. Based on the actual outline, spatial location information and defect distribution map of the concrete component, the surface of the component is divided into several grid units according to the three-dimensional coordinate range; Set planning constraints for detection points, construct an objective function, use a genetic algorithm to solve for the optimal solution, and output the three-dimensional coordinates of the detection points and the detection order; Real-time acquisition of on-site environmental interference parameters, and calculation of the surface flatness of concrete components based on captured images of concrete components; An environmental interference error database is constructed based on massive historical data. Compensation coefficients are obtained by matching real-time environmental parameters and correcting the preliminary results to obtain the final compressive strength value.

8. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 7, characterized in that, The process of obtaining the final compressive strength value specifically includes: Acquire massive amounts of historical detection data and use this data to construct a database of the correspondence between environmental interference errors; The real-time collected on-site environmental interference parameters are input into the database for matching to obtain the corresponding error compensation coefficients; The preliminary compressive strength test results are retrieved and secondary compensation corrections are performed to eliminate the influence of environmental factors and component surface condition on the test results, thus obtaining the final compressive strength test results of the concrete component.

9. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 1, characterized in that, The intensity dynamic prediction and tracing of S6 specifically includes: Based on acoustic emission data, the acoustic emission event density per unit time is calculated, whereby the event density is the number of acoustic emission events detected within a unit time window; the average event energy, average amplitude, and proportion of high-frequency components are statistically analyzed to characterize the activity level of internal cracks in the component; based on micro-vibration detection data, the rate of change of the natural frequency of micro-vibration, the rate of change of damping ratio, and the change in mode shape correlation coefficient are calculated within adjacent monitoring periods. By jointly mapping the acoustic emission event density of the excitation device with the rate of change of the natural frequency of micro-vibration, an internal damage index is formed. The internal damage index is expressed as: Internal Damage Index Acoustic emission event density normalization value Natural frequency relative rate of change Rate of change of damping ratio, where The weighting coefficients are determined based on historical test data or engineering calibration; damage-sensitive information from different sources is uniformly quantified into a single time series through internal damage indicators, and an internal damage time series curve is constructed. In the initial stage, the current compressive strength value of the component is obtained through a non-destructive testing fusion model. In non-critical parts or during the test stage, some true strength values ​​are obtained as calibration benchmarks. Compressive strength prediction is implemented. Historical strength data, internal damage indicators and environmental impact factors are used as input variables. The predicted strength values ​​at multiple future times are output. The prediction results form a compressive strength evolution curve with time as the horizontal axis and compressive strength as the vertical axis. Correlation analysis is performed on the internal damage time series curve and the compressive strength evolution curve. By calculating the correlation coefficient, lag correlation and trend consistency between the two, the influence law of internal damage changes on strength decay is identified. When the system detects an abnormal increase in the internal damage index in a short period of time, and its growth trend is highly correlated with the downward trend of the strength prediction curve, it is judged as a strength deterioration risk signal, and an early warning of component strength deterioration risk is issued.

10. The method for testing the compressive strength of building concrete components based on multi-technology integration according to claim 9, characterized in that, The deep learning feature fusion model constructed by S3 adopts a two-layer network structure of "base model" and "matching transfer layer", wherein, The base model is responsible for learning the general feature-intensity mapping relationship, and the proportioning transfer layer dynamically corrects the output of the base model according to the input proportioning parameters to achieve adaptive adjustment under different material systems. The mix transfer layer is responsible for automatically entering the online fine-tuning mode when the system detects that the concrete components of the project use a new mix ratio not covered by the database. It uses a small number of samples to quickly train the transfer layer, enabling the deep learning feature fusion model to converge quickly under the new mix ratio conditions. By embedding a state evolution sub-model into the deep learning feature fusion model, the time information of concrete from the initial setting period, early strength development period to long-term service period is transformed into learnable parameters and used for the learning and training of the state evolution sub-model, enabling the state evolution sub-model to dynamically adjust the prediction logic as the concrete age changes.

Images