Shrinkage test method of ultra-high performance concrete

Abstract

The invention discloses an ultra-high performance concrete shrinkage test method, and relates to the technical field of concrete monitoring, and the method comprises the following steps: recording dynamic changes of an internal pore structure and moisture distribution of concrete in real time, and generating a real-time pore distribution curve and a moisture change curve through a spectral image processing algorithm; constructing an incremental compensation model based on the pore distribution curve and the moisture change curve, and calculating a dynamic compensation factor by using a Kalman filtering algorithm; inputting the dynamic monitoring data packet into the incremental compensation model, updating the incremental compensation model by combining the dynamic compensation factor, and generating a contraction behavior prediction curve; and performing fitting analysis on the contraction behavior prediction curve and the dynamic monitoring data packet, extracting a key contraction rule, and generating a rule analysis report. According to the method, the key shrinkage rule is extracted through multi-step prediction, the rule analysis report is output, the method has remarkable advantages in the aspects of high-precision prediction, real-time monitoring and rule extraction, and optimization of concrete material performance and improvement of engineering quality are facilitated.

Description

Technical Field

[0001] The invention relates to the technical field of concrete monitoring, in particular to a shrinkage testing method for ultra-high performance concrete. Background Art

[0002] With the continuous improvement of concrete performance requirements in construction projects, ultra-high performance concrete (UHPC) has been widely used in many fields due to its excellent strength and durability. However, the shrinkage behavior of UHPC is more complicated than that of traditional concrete. Especially at the early age stage, the cracking problem caused by shrinkage may seriously affect its service performance. Therefore, studying and predicting the shrinkage behavior of UHPC has become the key to improving its long-term performance. In the existing technology, static testing methods or dynamic monitoring methods based on a single parameter are mostly used to study the shrinkage of UHPC. However, these methods are difficult to fully reflect the dynamic effects of porosity and moisture changes on shrinkage behavior, resulting in insufficient shrinkage prediction accuracy.

[0003] The shortcomings of existing technologies are mainly reflected in the following aspects: First, the dynamic coupling effect of pore evolution and water migration is not combined. Existing methods usually ignore the relationship between the two, resulting in inaccurate model prediction results; second, most existing models are static or semi-dynamic models, lacking real-time correction mechanisms, and are difficult to effectively respond to dynamic changes in complex environments. Therefore, a test method that can collect and analyze the dynamic changes of pores and water in real time is urgently needed to improve the accuracy of UHPC shrinkage behavior prediction. Summary of the invention

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

[0005] Therefore, the present invention provides a shrinkage test method for ultra-high performance concrete to solve the problems in the prior art that the model prediction results are not accurate enough and it is difficult to effectively cope with dynamic changes in complex environments.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a shrinkage test method for ultra-high performance concrete, which comprises: Real-time record of dynamic changes in the pore structure and moisture distribution inside the concrete, and generation of real-time pore distribution curves and moisture change curves through spectral image processing algorithms; An incremental compensation model is constructed based on the pore distribution curve and the moisture change curve, and the dynamic compensation factor is calculated using the Kalman filter algorithm. Using dynamic compensation factors, the contribution ratio of moisture gradient and pore change to shrinkage behavior in the incremental compensation model is adjusted in real time; During the dynamic compensation factor adjustment process, real-time pore distribution data and moisture change data are continuously collected to form a dynamic monitoring data package; Inputting the dynamic monitoring data packet into the incremental compensation model, updating the incremental compensation model in combination with the dynamic compensation factor, and generating a shrinkage behavior prediction curve; Fit and analyze the contraction behavior prediction curve and dynamic monitoring data package, extract key contraction rules and generate a rule analysis report.

[0007] As a preferred solution of the shrinkage test method of ultra-high performance concrete of the present invention, wherein: the real-time recording of the dynamic changes of the pore structure and moisture distribution inside the concrete, and the generation of the real-time pore distribution curve and moisture change curve by the spectral image processing algorithm, the specific steps are as follows: A high-frequency spectral imaging device is used to scan concrete samples point by point on a fixed grid point array, and multi-band spectral images are collected to generate time-series spectral data; Perform noise removal and normalization on the time series spectral data to generate a spectral information sequence; Based on the spectral information sequence, the porosity and moisture content are calculated through the characteristic bands and integrated to generate the pore moisture characteristic matrix; Based on the pore water feature matrix, the pore area is identified through image segmentation algorithm, and the porosity changes of each sampling point in time and depth dimensions are output; According to the porosity change, the dynamic evolution process of concrete pores is reconstructed layer by layer to generate a time series pore distribution map; The moisture content at different sampling points is calculated by combining the characteristic band ratio method to generate a moisture distribution map; The time series statistics of the time series pore distribution diagram and the moisture distribution diagram are performed to generate the pore distribution curve and the moisture change curve.

[0008] As a preferred solution of the shrinkage test method of ultra-high performance concrete of the present invention, wherein: the incremental compensation model is constructed based on the pore distribution curve and the moisture change curve, and the dynamic compensation factor is calculated using the Kalman filter algorithm. The specific steps are as follows: Based on the pore distribution curve, the porosity difference between adjacent time points is calculated by differential calculation method to obtain the porosity change rate ; Based on the moisture change curve, the moisture gradient is obtained by calculating the moisture change rate along the depth direction. ; The shrinkage behavior of concrete is generated by linearly increasing the porosity change rate and the moisture gradient. Based on the shrinkage behavior of concrete and the pore water characteristic matrix, a dynamic compensation factor is introduced The shrinkage behavior of different features is corrected and an incremental compensation model is constructed. The expression is: ; in, Indicates time The concrete shrinkage measured at any time, Indicates time The concrete shrinkage measured at any time, It represents the time difference between two consecutive acquisitions of concrete pore distribution and moisture distribution data. represents the fixed influence weight of pore change on shrinkage behavior, represents the fixed influence weight of moisture gradient on shrinkage behavior; Using Kalman filtering to calculate the dynamic compensation factor , the Kalman filter update expression is: ; ; in, Indicates time The dynamic compensation factor at the moment, Indicates at time The Kalman gain at time t, Indicates at time The prediction error at time Indicates at time The actual contraction behavior value observed at the moment, Indicates at time The contraction behavior value predicted by the incremental compensation model at this moment; The Kalman gain is calculated based on the ratio of the process error covariance to the observation noise covariance at the current moment. The expression is: ; in, represents the process error covariance, represents the observation noise covariance; Based on the Kalman gain, the process error covariance is updated at each time step, and the expression is: ; in, Indicates time The process error covariance at time , Indicates the proportion of the prediction error covariance retained at the previous moment, represents the process noise covariance.

[0009] As a preferred solution of the shrinkage test method of ultra-high performance concrete of the present invention, the dynamic compensation factor is used to adjust the contribution ratio of the moisture gradient and pore change in the incremental compensation model to the shrinkage behavior in real time. The specific steps are as follows: Perform principal component analysis on the pore water characteristic matrix, solve the covariance eigenvalue and covariance eigenvector of the covariance matrix, and form an eigenvalue set; The principal components of the eigenvalues ​​in the eigenvalue set are extracted again through principal component analysis, and the variance contribution rate of each principal component is calculated as the initial contribution weight; Based on the dynamic compensation factor calculated by Kalman filter, the initial contribution weight is dynamically corrected; The contribution ratio of moisture gradient and pore change in the incremental compensation model is updated using the dynamically corrected initial contribution weights.

[0010] As a preferred solution of the shrinkage test method of ultra-high performance concrete of the present invention, in which: during the dynamic compensation factor adjustment process, real-time pore distribution data and moisture change data are continuously collected to form a dynamic monitoring data packet. The specific steps are as follows: Before each dynamic compensation factor update, the pore image and moisture data at the current moment are collected in real time; Convolutional neural network is used to extract pore features in pore images, and multi-layer perceptron is used to extract moisture features in moisture data; Based on the fusion model, pore characteristics and water characteristics are integrated to form a fusion characteristic matrix; The time series input is constructed using the fused feature matrix, and the short-term contraction behavior prediction value of the future time step is generated through the long short-term memory network; Calculate the prediction error between the predicted value and the actual measured value of short-term contraction behavior, dynamically adjust the feature weights through an adaptive feedback mechanism, and optimize the fusion feature matrix; The fused feature matrix, short-term contraction behavior prediction value and prediction error are integrated according to timestamp to generate a dynamic monitoring data packet.

[0011] As a preferred solution of the shrinkage test method of ultra-high performance concrete of the present invention, wherein: the dynamic monitoring data packet is input into the incremental compensation model, the incremental compensation model is updated in combination with the dynamic compensation factor, and the shrinkage behavior prediction curve is generated. The specific steps are as follows: The time series of dynamic monitoring data packets are aligned by interpolation alignment method to form the input data of incremental compensation model; The extracted porosity change rate and moisture gradient are spliced ​​with the fused feature matrix to generate an enhanced feature matrix; The dynamic feature matrix is ​​generated by weighted correction of the enhanced feature matrix according to the dynamic compensation factor calculated by Kalman filtering; Using the dynamic feature matrix to update the incremental compensation model and calculate the predicted value of the contraction behavior at the current moment; Use the long short-term memory network to make short-term predictions on the dynamic feature matrix to generate predicted values ​​of contraction behavior in future time steps; The multi-step prediction results are combined to generate a continuous shrinkage behavior prediction curve.

[0012] As a preferred solution of the shrinkage test method of ultra-high performance concrete of the present invention, the shrinkage behavior prediction curve and the dynamic monitoring data package are fitted and analyzed to extract the key shrinkage law and generate a law analysis report. The specific steps are as follows: The contraction behavior prediction curve is aligned with the short-term contraction behavior prediction value in the dynamic monitoring data package in time series to form a fitting data set; The least square method is used to perform error analysis on the fitting data set and calculate the fitting error curve; Smooth the fitting error curve and extract the error change trend characteristics; Identify key contraction stages based on error change trend characteristics and extract stage rules; The characteristic contributions of the phased regularities are ranked and the main characteristic regularities are extracted through principal component analysis; Generate a key shrinkage law analysis report based on the main feature law and output the law chart to generate a law analysis report.

[0013] As a preferred embodiment of the shrinkage test method for ultra-high performance concrete of the present invention, the least square method is used to perform error analysis on the fitting data set to calculate the fitting error curve. The specific steps are as follows: Python+NumPy is used to calculate the error of each time point in the fitting data set point by point and output the error sequence; Apply the least square method to perform polynomial fitting on the error sequence to generate a fitting error curve; The fitting error curve is smoothed by Savitzky-Golay filtering, and the fitting error curve that changes with time is output.

[0014] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the shrinkage test method for ultra-high performance concrete as described in the first aspect of the present invention is implemented.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the shrinkage test method for ultra-high performance concrete as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: by real-time recording of pore distribution and moisture changes, combined with high-frequency spectral imaging and multi-band feature extraction methods, the dynamic evolution characteristics inside the concrete can be effectively captured to generate high-precision time series data; the Kalman filter algorithm is introduced to calculate the dynamic compensation factor, and the incremental compensation model can be dynamically corrected at different time steps, significantly reducing the shrinkage prediction error; in addition, the feature extraction and prediction mechanism of the fusion convolutional neural network and the long short-term memory network can monitor the shrinkage behavior of concrete in real time, and can also extract key shrinkage laws through multi-step prediction and output a law analysis report, which has significant advantages in high-precision prediction, real-time monitoring and law extraction, and helps to optimize the performance of concrete materials and improve engineering quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 Flow chart of the shrinkage test method of ultra-high performance concrete in Example 1.

[0019] Figure 2 Flow chart of the method for calculating the fitting error curve in Example 1. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, reference Figure 1 and Figure 2, which is the first embodiment of the present invention, provides a shrinkage test method for ultra-high performance concrete, comprising the following steps: S1. Real-time record of dynamic changes in the pore structure and moisture distribution inside the concrete, and generation of real-time pore distribution curves and moisture change curves through spectral image processing algorithms.

[0024] Furthermore, a high-frequency spectral imaging device is used to scan the concrete sample point by point on a fixed grid point array, and collect multi-band spectral images to generate time-series spectral data; Perform noise removal and normalization on the time series spectral data to generate a spectral information sequence; It should be noted that the noise removal and normalization processing specifically uses the wavelet denoising method to eliminate high-frequency noise interference, and uses the maximum and minimum normalization or standard normal normalization method to scale the spectral data.

[0025] Based on the spectral information sequence, the porosity and moisture content are calculated through the characteristic bands and integrated to generate the pore moisture characteristic matrix; It should be noted that the characteristic band refers to the specific wavelength region in the spectrum that is closely related to the porosity and moisture content, such as the two main absorption peaks of moisture in the near-infrared spectrum at 970nm and 1450nm. The changes in the reflection intensity of these bands directly reflect the moisture content in the concrete, while the porosity is indirectly characterized by the scattering characteristics of light between media of different densities. Based on the spectral information sequence, the spectral intensity of each sampling point is extracted within the characteristic band range, and specific physical models or empirical formulas are used, such as the Beer-Lambert law combined with the ratio method and the scattering characteristic model, to calculate the porosity and moisture content respectively. The moisture content can be calculated by the ratio method, that is, the reflectance ratio of the band corresponding to the moisture absorption peak and the reference band is selected, and the moisture content is calculated according to the pre-calibrated relationship between absorbance and content; the porosity is calculated by the association model of the spectral reflectance intensity and the known porosity sample. The association model is a random forest regression model. Finally, the porosity and moisture content of each sampling point at different time and depth are integrated to generate a pore moisture characteristic matrix with spatiotemporal characteristics. Each row of the pore moisture characteristic matrix represents the porosity and moisture content of a sampling point at a specific time point, which serves as the basis for subsequent dynamic identification.

[0026] Based on the pore water feature matrix, the pore area is identified through image segmentation algorithm, and the porosity changes of each sampling point in time and depth dimensions are output; Specifically, based on the spectral information sequence, 970nm and 1450nm in the near-infrared band are first selected as characteristic bands, where the 970nm band corresponds to the strong absorption peak of water, and the 1450nm band reflects the change in the gas phase content in the pores, ensuring that the distribution characteristics of pores and water can be characterized at the same time; then, by calculating the ratio of the reflected light intensity to the reference light intensity under the characteristic band, the porosity and water content of each sampling point are obtained according to the Beer-Lambert law, and then the porosity and water content of all sampling points in the time and depth dimensions are integrated to generate a pore water feature matrix containing spatial and temporal characteristics. Then, using the pore water feature matrix as input, image segmentation algorithms such as convolutional neural networks (CNNs) are used to accurately identify the pore area, and the porosity changes of each sampling point are calculated through the recognition results at different time points and depth levels. Finally, the porosity changes of each sampling point in the time and depth dimensions are output, providing high-precision basic data for subsequent dynamic evolution modeling.

[0027] According to the porosity change, the dynamic evolution process of concrete pores is reconstructed layer by layer to generate a time series pore distribution map; Specifically, according to the porosity change, the porosity of sampling points at different depths at each time point is first interpolated to ensure the continuity of the porosity in space, and then the interpolated data is accumulated layer by layer along the depth direction to reconstruct a complete three-dimensional pore distribution model at each time point; then, the three-dimensional pore model of consecutive time points is serialized to identify dynamic change characteristics such as pore expansion, contraction or closure, and the change rate of the porosity of each layer is calculated. Finally, the three-dimensional pore distribution at each time point is arranged in sequence to generate a complete time series pore distribution map, which clearly shows the dynamic evolution process of the pores in time and space.

[0028] The moisture content at different sampling points is calculated by combining the characteristic band ratio method to generate a moisture distribution map; The time series statistics of the time series pore distribution diagram and the moisture distribution diagram are performed to generate the pore distribution curve and the moisture change curve.

[0029] Specifically, when performing time-series statistics on the time-series pore distribution map and moisture distribution map, first extract the porosity change value of the pore distribution data at continuous time points at fixed time intervals, and perform spatial weighted average of the porosity of each sampling point to obtain the overall porosity sequence that changes with time in the entire sample area; then, use the same method for the moisture content data in the moisture distribution map to extract the moisture content change value at each time point, and perform weighted average of all sampling points to obtain the overall moisture change sequence. Finally, the extracted overall porosity sequence and moisture change sequence are plotted into pore distribution curves and moisture change curves along the time axis for subsequent dynamic monitoring and construction of incremental compensation models.

[0030] S2. An incremental compensation model is constructed based on the pore distribution curve and the moisture change curve, and the dynamic compensation factor is calculated using the Kalman filtering algorithm.

[0031] Furthermore, based on the pore distribution curve, the porosity difference between adjacent time points is calculated by differential calculation method to obtain the porosity change rate ; The evolution of pore structure plays a key role in the shrinkage of concrete, and This effectively quantifies the change. The introduction of enables the incremental compensation model to capture the microscopic change characteristics of the internal structure of concrete in real time, thereby improving the accuracy of shrinkage behavior prediction.

[0032] Based on the moisture change curve, the moisture gradient is obtained by calculating the moisture change rate along the depth direction. ; It should be noted that the shrinkage evolution of concrete at different time points can be accurately captured through the microscopic change characteristics of pores and moisture. Compared with the traditional method based only on total shrinkage, and The introduction of provides more fine-grained data and significantly improves the prediction accuracy.

[0033] The shrinkage behavior of concrete is generated by linearly increasing the porosity change rate and the moisture gradient. Based on the shrinkage behavior of concrete and the pore water characteristic matrix, a dynamic compensation factor is introduced The shrinkage behavior of different features is corrected and an incremental compensation model is constructed. The expression is: ; in, Indicates time The concrete shrinkage measured at any time, Indicates time The concrete shrinkage measured at any time, It represents the time difference between two consecutive acquisitions of concrete pore distribution and moisture distribution data. represents the fixed influence weight of pore change on shrinkage behavior, represents the fixed influence weight of moisture gradient on shrinkage behavior; Preferably, the incremental compensation model can dynamically correct the predicted value of shrinkage behavior at each time step, ensuring that the prediction results remain highly accurate over time. Compared with static models, incremental compensation models are more adaptable to dynamic changes in the environment. Traditional models cannot adjust the weights of different factors in real time, such as macroscopic shrinkage prediction models based on empirical formulas (such as ACI, CEB-FIP models) and static linear regression models. These traditional models usually use fixed weights or parameters, which are difficult to respond to dynamic changes in the environment and material properties in real time. The incremental compensation model solves the problem of slow response of static models to environmental changes through dynamic weight correction.

[0034] It should be noted that and The determination of the two weights needs to be based on a method that combines experimental data analysis with theoretical models. The specific steps are as follows: First, a series of shrinkage test experiments are conducted on different types of concrete samples to record the change data of porosity and moisture gradient in real time, and measure the corresponding shrinkage; then, the collected data are subjected to linear regression analysis using the least squares method, with the porosity change rate and moisture gradient as independent variables and the actually measured shrinkage as the dependent variable, to solve the linear coefficients in the regression equation, thereby obtaining and In this process, in order to ensure the reliability and universality of the experimental results, it is necessary to statistically analyze the experimental data under multiple sets of different environmental conditions (such as humidity, temperature, etc.), and take the average of the coefficients of all regression equations to finally determine the and The fixed impact weights.

[0035] Furthermore, and To verify the rationality and stability of the incremental compensation model, after the weights are initially determined, they need to be substituted into the incremental compensation model for multiple predictions and fitting analysis of the actual measured values. Specifically, by introducing dynamic compensation factors and combining real-time data at different time steps, the error between the predicted value of the shrinkage behavior of the incremental compensation model and the actual measured value is calculated, and the error is continuously adjusted with the goal of minimizing the error. and The initial value is changed until the error is stable and meets the accuracy requirement. For example, when the absolute value of the error or the mean square error changes less than the set convergence standard (for example, less than 0.001) in multiple consecutive iterations, and the mean square error is lower than the target accuracy threshold (for example, less than 0.01), it is considered to meet the accuracy requirement. The final weight value is the optimal fixed influence weight under the current concrete type and environmental conditions, ensuring that the incremental compensation model has high prediction accuracy and adaptability under different conditions.

[0036] Using Kalman filtering to calculate the dynamic compensation factor , the Kalman filter update expression is: ; ; in, Indicates time The dynamic compensation factor at the moment, Indicates at time The Kalman gain at time t, Indicates at time The prediction error at time Indicates at time The actual contraction behavior value observed at the moment, Indicates at time The contraction behavior value predicted by the incremental compensation model at this moment; It should be noted that the dynamic compensation factor is updated based on the Kalman filter, which can realize the dynamic correction of the error and ensure the stability of the incremental compensation model in long-term operation. At the same time, the dynamic compensation factor is reasonable, because the shrinkage behavior of concrete is affected by many factors such as the environment and material properties. Correcting the dynamic compensation factor through Kalman filtering can significantly reduce the prediction error.

[0037] Preferably, the Kalman filter mechanism can dynamically adjust the compensation factor according to the real-time error feedback, thereby maintaining the accuracy of the incremental compensation model under different environmental conditions and avoiding the failure of the incremental compensation model due to environmental changes. Existing methods usually use fixed compensation factors, which are difficult to cope with dynamic changes in complex environments. This step dynamically adjusts the compensation factor through Kalman filtering, significantly enhancing the adaptability of the incremental compensation model.

[0038] Through the Kalman gain, based on the sensitivity of the error correction at the current moment, the expression is: ; in, represents the process error covariance, represents the observation noise covariance; Preferably, Kalman gain is the core parameter of Kalman filtering, which is used to weigh the credibility between the predicted value and the observed value of the contraction behavior.

[0039] The process error covariance is updated at each time step, expressed as: ; in, Indicates time The process error covariance at time , Indicates the proportion of the prediction error covariance retained at the previous moment, represents the process noise covariance.

[0040] It should be noted that by introducing dynamic compensation factors, Kalman filtering and incremental compensation models, the shortcomings of existing technologies in dynamic adaptability, accuracy and real-time performance are solved, and significant advantages can be achieved in the field of concrete shrinkage prediction and compensation.

[0041] S3. Use the dynamic compensation factor to adjust the contribution ratio of moisture gradient and pore change to shrinkage behavior in the incremental compensation model in real time.

[0042] Furthermore, principal component analysis is performed on the pore water characteristic matrix to solve the covariance eigenvalues ​​and covariance eigenvectors of the covariance matrix to form an eigenvalue set; It should be noted that principal component analysis (PCA) is used to reduce feature dimensions while retaining the maximum amount of information. By calculating the variance contribution of each principal component, the relative importance of pore changes and moisture gradients in the overall shrinkage behavior can be determined. Optimally, the dynamic physical properties are converted into quantifiable statistical features, effectively reducing the complexity of the incremental compensation model.

[0043] The principal components of the eigenvalues ​​in the eigenvalue set are extracted again through principal component analysis, and the variance contribution rate of each principal component is calculated as the initial contribution weight; Based on the dynamic compensation factor calculated by Kalman filter, the initial contribution weight is dynamically corrected; Specifically, in the process of calculating the dynamic compensation factor based on Kalman filtering, the real-time measured porosity change rate and moisture gradient are first extracted and compared with the predicted value of the shrinkage behavior of the incremental compensation model at the current moment, so as to calculate the error between the actual observation value and the predicted value of the shrinkage behavior. The error is then used to calculate the Kalman gain, which is used to measure the influence weight of the current observation value on the correction of the incremental compensation model. The calculation process involves the covariance of the process error and the observation noise at the current moment. After obtaining the Kalman gain, the dynamic compensation factor is corrected by the gain according to the compensation factor update formula so that it can respond to the dynamic changes of the external environment in a timely manner. The corrected dynamic compensation factor is directly used to adjust the initial weight, so that the contribution ratio of moisture gradient and porosity change to the shrinkage behavior can be dynamically updated in each time step, thereby realizing the continuous optimization of the incremental compensation model and improving the prediction accuracy and stability of the incremental compensation model in a dynamic environment.

[0044] The contribution ratio of moisture gradient and pore change in the incremental compensation model is updated using the dynamically corrected initial contribution weights.

[0045] It should be noted that the corrected initial contribution weights are used to update the incremental compensation model in real time, so that it can more accurately reflect the contribution of pore changes and moisture gradients to shrinkage at the current moment. This dynamic update mechanism greatly improves the real-time and accuracy of the incremental compensation model, especially in practical engineering applications, it can provide more reliable predictions of shrinkage behavior. Traditional static models often fail in engineering due to environmental changes. The present invention dynamically updates the incremental compensation model to make the prediction results more accurate and reliable, solving the problem of insufficient applicability of traditional methods.

[0046] S4. During the dynamic compensation factor adjustment process, real-time pore distribution data and moisture change data are continuously collected to form a dynamic monitoring data package.

[0047] Furthermore, before each dynamic compensation factor is updated, the pore image and moisture data at the current moment are collected in real time; Convolutional neural network is used to extract pore features in pore images, and multi-layer perceptron is used to extract moisture features in moisture data; It should be noted that the pore features are mainly extracted by convolutional neural networks (CNNs), which utilizes the local perception ability of CNNs in image feature extraction, enabling them to capture texture details and geometric characteristics in pore images. The moisture features are extracted by multi-layer perceptrons (MLPs), because MLPs have the advantages of high efficiency and good feature extraction accuracy in processing one-dimensional data (such as sequences of moisture content).

[0048] Compared with traditional feature extraction methods, such as manually defined feature extraction, CNN and MLP can automatically learn deep features in images and numerical sequences, thereby reducing errors that may be caused by human intervention. This automatic feature learning method can significantly improve the accuracy and consistency of feature extraction, especially when dealing with complex structures (such as concrete pore morphology), and its effect is far better than traditional methods.

[0049] Based on the fusion model, pore characteristics and water characteristics are integrated to form a fusion characteristic matrix; Preferably, the fusion model is used to integrate the pore characteristics and moisture characteristics, and the resulting fusion feature matrix contains both the spatial information of concrete pore changes and the temporal information of moisture dynamic changes. This fusion method avoids the deviation caused by a single feature and improves the comprehensiveness of the prediction.

[0050] The time series input is constructed using the fused feature matrix, and the short-term contraction behavior prediction value of the future time step is generated through the long short-term memory network; Specifically, the extracted pore characteristics and moisture characteristics are arranged in chronological order to form a time series data sequence containing multi-dimensional features, and then the fixed-length time segments are divided as input sequences through the sliding window method to ensure that each time segment contains enough information to capture short-term dynamic change trends. Next, these time series are input into the long short-term memory network (LSTM). The LSTM unit effectively retains the key time-dependent characteristics and filters out irrelevant noise information through the synergy of the internal memory state and the forget gate, input gate and output gate, and finally outputs the short-term shrinkage behavior prediction value of the future time step.

[0051] Furthermore, in order to improve the accuracy of the prediction, the LSTM network will continuously adjust the weights through the back-propagation algorithm during the training process, so that the predicted value of the contraction behavior gradually approaches the actual contraction behavior.

[0052] Calculate the prediction error between the predicted value and the actual measured value of short-term contraction behavior, dynamically adjust the feature weights through an adaptive feedback mechanism, and optimize the fusion feature matrix; Specifically, first, the generated short-term shrinkage behavior prediction value is compared point by point with the actual measurement value obtained by real-time measurement, and the error is calculated to form an error sequence. Next, the change trend of the error sequence is analyzed through the adaptive feedback mechanism, and the weights of different features in the fusion feature matrix are automatically adjusted. The core of the adaptive feedback mechanism is to adjust the weights of each feature in the fusion feature matrix in real time according to the dynamic changes of the prediction error. Through this mechanism, the key features that significantly affect the shrinkage behavior of concrete (such as porosity change rate and moisture gradient) will be given higher weights, thereby increasing their contribution in the incremental compensation model; while the secondary features that have less impact on the shrinkage behavior (such as the principal component with lower variance contribution) are appropriately weakened to reduce their interference. The optimized fusion feature matrix will be used as input to re-participate in the prediction calculation of the incremental compensation model and the long short-term memory network (LSTM), ensuring that the predicted incremental compensation model can continuously adapt to the dynamic changes of the external environment, thereby improving the prediction accuracy of the shrinkage behavior and the robustness of the incremental compensation model.

[0053] The fused feature matrix, short-term contraction behavior prediction value and prediction error are integrated according to timestamp to generate a dynamic monitoring data packet.

[0054] It should be noted that by collecting pore images and moisture data in real time, using deep learning models to extract features and fuse them into time series inputs, high-precision prediction of concrete shrinkage behavior can be achieved. Traditional concrete shrinkage prediction methods mostly rely on empirical models and linear assumptions, which are difficult to accurately reflect dynamic changes in complex environments. Preferably, by introducing deep learning algorithms such as convolutional neural networks and long short-term memory networks, combined with real-time collected dynamic data, the limitations of traditional methods on nonlinear dynamic characteristics have been broken through.

[0055] S5. Input the dynamic monitoring data packet into the incremental compensation model, update the incremental compensation model in combination with the dynamic compensation factor, and generate a shrinkage behavior prediction curve.

[0056] Extract the porosity change rate, moisture gradient and fusion feature matrix from the dynamic monitoring data package, align them in time series to form the input data of the incremental compensation model; Preferably, by extracting key features and aligning them in time series, the temporal consistency of the input data can be ensured, and the responsiveness of the incremental compensation model to short-term dynamic changes can be improved. This data preprocessing method can solve the problem of ignoring time correlation in traditional static models.

[0057] The extracted porosity change rate and moisture gradient are spliced ​​with the fused feature matrix to generate an enhanced feature matrix; The dynamic feature matrix is ​​generated by weighted correction of the enhanced feature matrix according to the dynamic compensation factor calculated by Kalman filtering; Specifically, first, the porosity change rate, moisture gradient and fusion feature matrix in the dynamic monitoring data packet are extracted to generate an enhanced feature matrix; then, the dynamic compensation factor at the current moment is calculated using the Kalman filter algorithm. Subsequently, each feature in the enhanced feature matrix is ​​weighted and corrected according to the dynamic compensation factor. By assigning different time-varying weights to the porosity change rate and moisture gradient, a dynamic feature matrix that can more accurately characterize the dynamic changes is generated. This correction process ensures the adaptability and accuracy of the incremental compensation model under complex conditions.

[0058] Using the dynamic feature matrix to update the incremental compensation model and calculate the predicted value of the contraction behavior at the current moment; Specifically, the dynamic feature matrix generated in the previous step is input into the incremental compensation model, and according to the time-varying characteristics of the porosity change rate and moisture gradient in the dynamic feature matrix, it is updated in real time through the modified linear incremental expression to obtain the predicted value of the shrinkage behavior at the current moment; in this process, the dynamic compensation factor calculated by the Kalman filter continuously corrects the weights of the input features of the incremental compensation model to ensure that the prediction results are continuously optimized as the environment changes; through this dynamic update mechanism, the incremental compensation model can reflect the actual shrinkage trend of concrete in real time, significantly improving the real-time and accuracy of the prediction.

[0059] Use the long short-term memory network to make short-term predictions on the dynamic feature matrix to generate predicted values ​​of contraction behavior in future time steps; Specifically, the dynamic feature matrix is ​​input into the long short-term memory network (LSTM) in time series. The network captures the long-term and short-term dependencies of the time series in the dynamic feature matrix through memory gating units, so as to accurately identify the evolution pattern of contraction behavior. The LSTM model generates contraction behavior prediction values ​​in several future time steps by recursively processing the features of continuous time steps. Subsequently, the contraction behavior prediction value at the current moment is combined with the prediction results of future time steps to generate a continuous contraction prediction sequence in the short term.

[0060] It should be noted that this short-term prediction method based on the long short-term memory network (LSTM) not only effectively solves the problem that traditional linear models are difficult to capture nonlinear dynamic changes, but also provides a decision-making basis for taking compensatory measures in advance in practical applications.

[0061] The multi-step prediction results are combined to generate a continuous shrinkage behavior prediction curve.

[0062] The best method is to introduce dynamic compensation factors, fusion feature matrix and long short-term memory network prediction model to achieve high-precision dynamic prediction of concrete shrinkage behavior. The key steps not only improve the response speed and prediction accuracy of the incremental compensation model, but also effectively solve the problem of mismatch between dynamic changes of data and static characteristics of the model in traditional methods, thus having significant advantages in practical applications.

[0063] S6. Perform fitting analysis on the contraction behavior prediction curve and the dynamic monitoring data package, extract key contraction rules and generate a rule analysis report.

[0064] Furthermore, the contraction behavior prediction curve is aligned with the actual contraction data in the dynamic monitoring data packet in time series to form a fitting data set; The fitting data set refers to the data set that aligns the contraction behavior prediction curve with the actual contraction data in the dynamic monitoring data package in time series. Its role is to provide basic data for subsequent error analysis and ensure the accuracy of error calculation. The construction process of the fitting data set needs to ensure the synchronization of the time series to avoid errors introduced by inconsistent data time points.

[0065] It should be noted that error analysis by fitting the data set can accurately quantify the difference between the predicted value of contraction behavior and the actual measured value. This accurate quantification helps to locate the weak links in the prediction of the incremental compensation model, thereby improving the overall prediction performance of the incremental compensation model. Different from the prior art that relies only on a single data source for analysis, the present invention significantly enhances the credibility of the analysis results by introducing the alignment of the dynamic monitoring data packet with the contraction behavior prediction curve.

[0066] The least square method is used to perform error analysis on the fitting data set and calculate the fitting error curve; Specifically, Python+NumPy is used to calculate the error at each time point in the fitting data set point by point and output the error sequence; Apply the least square method to perform polynomial fitting on the error sequence to generate a fitting error curve; The fitting error curve is smoothed by Savitzky-Golay filtering, and the fitting error curve that changes with time is output.

[0067] Smooth the fitting error curve and extract the error change trend characteristics; Specifically, when smoothing the fitting error curve and extracting the error change trend characteristics, the sliding window method is first used to smooth the error curve. The width of the sliding window needs to be reasonably set according to the data acquisition frequency and the error fluctuation amplitude, in order to reduce the interference of local random noise on the overall trend. After the smoothing process is completed, by analyzing the change rate and change trend of the smoothed curve, the characteristic information of the error change over time is further extracted. These characteristics include the inflection point position of the error change, the speed of rise or fall, and the duration of the trend. The output of this process is an analysis data set containing trend characteristics, which can provide an effective basis for the identification of key contraction stages.

[0068] Identify key contraction stages based on error change trend characteristics and extract stage rules; The characteristic contributions of the phased regularities are ranked and the main characteristic regularities are extracted through principal component analysis; Specifically, when identifying the key contraction stage and extracting the stage rules based on the error change trend characteristics, the inflection point position in the error change trend characteristics is first analyzed, and the time interval where the inflection point is located is used as the possible stage dividing point. Subsequently, according to the speed and stability of the error change in each stage, the error curve is segmented and divided into several key contraction stages. On this basis, the error changes in each stage are statistically analyzed to extract the characteristic change rules within the stage, including the rate of change, the amplitude of change, and the trend of change. After identifying the stage characteristics, the rules of each stage are further modeled by the curve fitting method, so as to summarize the main characteristic rules of the key stages, which can provide a reliable basis and support for the subsequent analysis report.

[0069] Generate a key shrinkage law analysis report based on the main feature law and output the law chart to generate a law analysis report.

[0070] It should be noted that the generation of the law analysis report can not only intuitively display the key laws of concrete shrinkage, but also serve as the basis for subsequent construction monitoring and prediction, thus having important value in practical engineering applications. Different from the existing technology that only conducts local law analysis, the present invention can comprehensively reflect the overall trend and detailed changes of shrinkage behavior by combining multi-stage and multi-feature law extraction, greatly improving the engineering applicability of the analysis results.

[0071] This embodiment also provides a computer device, which is applicable to the shrinkage test method of ultra-high performance concrete, and includes: a memory and a processor. The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the shrinkage test method of ultra-high performance concrete proposed in the above embodiment.

[0072] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0073] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, the shrinkage test method for ultra-high performance concrete proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0074] In summary, the present invention can effectively capture the dynamic evolution characteristics inside concrete and generate high-precision time series data by real-time recording of pore distribution and moisture changes, combined with high-frequency spectral imaging and multi-band feature extraction methods; the Kalman filter algorithm is introduced to calculate the dynamic compensation factor, which can dynamically correct the incremental compensation model at different time steps, significantly reducing the shrinkage prediction error; in addition, the feature extraction and prediction mechanism of the fusion convolutional neural network and the long short-term memory network can monitor the shrinkage behavior of concrete in real time, and can also extract key shrinkage laws through multi-step prediction and output a law analysis report, which has significant advantages in high-precision prediction, real-time monitoring and law extraction, and helps to optimize the performance of concrete materials and improve engineering quality.

[0075] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A shrinkage test method for ultra-high performance concrete, characterized in that: include, Real-time record of dynamic changes in the pore structure and moisture distribution inside the concrete, and generation of real-time pore distribution curves and moisture change curves through spectral image processing algorithms; An incremental compensation model is constructed based on the pore distribution curve and the moisture change curve, and the dynamic compensation factor is calculated using the Kalman filter algorithm. Using dynamic compensation factors, the contribution ratio of moisture gradient and pore change to shrinkage behavior in the incremental compensation model is adjusted in real time; During the dynamic compensation factor adjustment process, real-time pore distribution data and moisture change data are continuously collected to form a dynamic monitoring data package; Inputting the dynamic monitoring data packet into the incremental compensation model, updating the incremental compensation model in combination with the dynamic compensation factor, and generating a shrinkage behavior prediction curve; Fit and analyze the contraction behavior prediction curve and dynamic monitoring data package, extract key contraction laws and generate law analysis reports.

2. The shrinkage test method for ultra-high performance concrete according to claim 1, characterized in that: The real-time recording of the dynamic changes of the pore structure and moisture distribution inside the concrete, and the generation of a real-time pore distribution curve and a moisture change curve through a spectral image processing algorithm, are specifically performed as follows: A high-frequency spectral imaging device is used to scan concrete samples point by point on a fixed grid point array, and multi-band spectral images are collected to generate time-series spectral data; Perform noise removal and normalization on the time series spectral data to generate a spectral information sequence; Based on the spectral information sequence, the porosity and moisture content are calculated through the characteristic bands and integrated to generate the pore moisture characteristic matrix; Based on the pore water feature matrix, the pore area is identified through image segmentation algorithm, and the porosity changes of each sampling point in time and depth dimensions are output; According to the porosity change, the dynamic evolution process of concrete pores is reconstructed layer by layer to generate a time series pore distribution map; The moisture content at different sampling points is calculated by combining the characteristic band ratio method to generate a moisture distribution map; The time series statistics of the time series pore distribution diagram and the moisture distribution diagram are performed to generate the pore distribution curve and the moisture change curve.

3. The shrinkage test method for ultra-high performance concrete according to claim 2, characterized in that: The incremental compensation model is constructed based on the pore distribution curve and the moisture change curve, and the dynamic compensation factor is calculated using the Kalman filter algorithm. The specific steps are as follows: Based on the pore distribution curve, the porosity difference between adjacent time points is calculated by differential calculation method to obtain the porosity change rate ; Based on the moisture change curve, the moisture gradient is obtained by calculating the moisture change rate along the depth direction. ; The shrinkage behavior of concrete is generated by linearly increasing the porosity change rate and the moisture gradient. Based on the shrinkage behavior of concrete and the pore water characteristic matrix, a dynamic compensation factor is introduced The shrinkage behavior of different features is corrected and an incremental compensation model is constructed. The expression is: ； in, Indicates time The concrete shrinkage measured at any time, Indicates time The concrete shrinkage measured at any time, It represents the time difference between two consecutive acquisitions of concrete pore distribution and moisture distribution data. represents the fixed influence weight of pore change on shrinkage behavior, represents the fixed effect weight of moisture gradient on shrinkage behavior; Using Kalman filtering to calculate the dynamic compensation factor , the Kalman filter update expression is: ； ； in, Indicates time The dynamic compensation factor at the moment, Indicates at time The Kalman gain at time t, Indicates at time The prediction error at time Indicates at time The actual contraction behavior value observed at the moment, Indicates at time The contraction behavior value predicted by the incremental compensation model at this moment; The Kalman gain is calculated based on the ratio of the process error covariance to the observation noise covariance at the current moment. The expression is: ； in, represents the process error covariance, represents the observation noise covariance; Based on the Kalman gain, the process error covariance is updated at each time step, and the expression is: ； in, Indicates time The process error covariance at time , Indicates the proportion of the prediction error covariance retained at the previous moment, represents the process noise covariance.

4. The shrinkage test method for ultra-high performance concrete according to claim 3, characterized in that: The dynamic compensation factor is used to adjust the contribution ratio of the moisture gradient and pore change to the shrinkage behavior in the incremental compensation model in real time. The specific steps are as follows: Perform principal component analysis on the pore water characteristic matrix, solve the covariance eigenvalue and covariance eigenvector of the covariance matrix, and form an eigenvalue set; The principal components of the eigenvalues ​​in the eigenvalue set are extracted again through principal component analysis, and the variance contribution rate of each principal component is calculated as the initial contribution weight; Based on the dynamic compensation factor calculated by Kalman filter, the initial contribution weight is dynamically corrected; The contribution ratio of moisture gradient and pore change in the incremental compensation model is updated using the dynamically corrected initial contribution weights.

5. The shrinkage test method for ultra-high performance concrete according to claim 4, characterized in that: In the process of adjusting the dynamic compensation factor, real-time pore distribution data and moisture change data are continuously collected to form a dynamic monitoring data package. The specific steps are as follows: Before each dynamic compensation factor update, the pore image and moisture data at the current moment are collected in real time; Convolutional neural network is used to extract pore features in pore images, and multi-layer perceptron is used to extract moisture features in moisture data; Based on the fusion model, pore characteristics and water characteristics are integrated to form a fusion characteristic matrix; The time series input is constructed using the fused feature matrix, and the short-term contraction behavior prediction value of the future time step is generated through the long short-term memory network; Calculate the prediction error between the predicted value and the actual measured value of short-term contraction behavior, dynamically adjust the feature weights through an adaptive feedback mechanism, and optimize the fusion feature matrix; The fused feature matrix, short-term contraction behavior prediction value and prediction error are integrated according to timestamp to generate a dynamic monitoring data packet.

6. The shrinkage test method for ultra-high performance concrete according to claim 5, characterized in that: The specific steps of inputting the dynamic monitoring data packet into the incremental compensation model, updating the incremental compensation model in combination with the dynamic compensation factor, and generating the contraction behavior prediction curve are as follows: The time series of dynamic monitoring data packets are aligned by interpolation alignment method to form the input data of incremental compensation model; The extracted porosity change rate and moisture gradient are spliced ​​with the fused feature matrix to generate an enhanced feature matrix; The dynamic feature matrix is ​​generated by weighted correction of the enhanced feature matrix according to the dynamic compensation factor calculated by Kalman filtering; Using the dynamic feature matrix to update the incremental compensation model and calculate the predicted value of the contraction behavior at the current moment; Use the long short-term memory network to make short-term predictions on the dynamic feature matrix to generate predicted values ​​of contraction behavior in future time steps; The multi-step prediction results are combined to generate a continuous shrinkage behavior prediction curve.

7. The shrinkage test method for ultra-high performance concrete according to claim 6, characterized in that: The specific steps of fitting and analyzing the contraction behavior prediction curve and the dynamic monitoring data package, extracting key contraction laws and generating a law analysis report are as follows: The contraction behavior prediction curve is aligned with the short-term contraction behavior prediction value in the dynamic monitoring data package in time series to form a fitting data set; The least square method is used to perform error analysis on the fitting data set and calculate the fitting error curve; Smooth the fitting error curve and extract the error change trend characteristics; Identify key contraction stages based on error change trend characteristics and extract stage rules; The characteristic contributions of the phased regularities are ranked and the main characteristic regularities are extracted through principal component analysis; Generate a key shrinkage law analysis report based on the main feature law and output the law chart to generate a law analysis report.

8. The shrinkage test method for ultra-high performance concrete according to claim 7, characterized in that: The least square method is used to perform error analysis on the fitting data set to calculate the fitting error curve. The specific steps are as follows: Python+NumPy is used to calculate the error of each time point in the fitting data set point by point and output the error sequence; Apply the least square method to perform polynomial fitting on the error sequence to generate a fitting error curve; The fitting error curve is smoothed by Savitzky-Golay filtering, and the fitting error curve that changes with time is output.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the shrinkage test method for ultra-high performance concrete according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the shrinkage test method for ultra-high performance concrete according to any one of claims 1 to 8 are implemented.

Images