Self-compacting concrete high temperature compressive strength rapid prediction method and traceability interpretation system

Abstract

A self-compacting concrete high-temperature compressive strength rapid prediction method and traceability interpretation system, by collecting existing self-compacting concrete high-temperature residual compressive strength research field mix proportion parameters, fiber parameters and fire characteristic parameter test data, training machine learning model, establishing a self-compacting concrete high-temperature compressive strength prediction model and traceability interpretation system, so that the self-compacting concrete high-temperature compressive strength rapid prediction can be realized, and more accurate, reliable and scientific intelligent decision support can be provided for post-disaster building repair; at the same time, based on the traceability interpretation of the self-compacting concrete high-temperature residual compressive strength rapid prediction model, engineering and technical personnel can know the core driving factors and influence degree of single variable characteristic parameter and its coupling effect on the self-compacting concrete high-temperature residual compressive strength loss, so that the related parameters of the self-compacting concrete can be optimized through the traceability interpretation result of the prediction model, and the high-temperature compressive strength of the self-compacting concrete can be improved.

Description

Technical Field

[0001] This invention relates to the field of post-fire engineering structure identification technology, specifically to a rapid prediction method and source tracing interpretation system for the high-temperature compressive strength of self-compacting concrete. Background Technology

[0002] Self-compacting concrete (SCC) is widely used in the construction of various civil and public buildings due to its excellent flowability and efficient construction. Both civil and public buildings are at risk of fire. In the event of a fire, it is necessary to promptly assess the affected structure and submit an assessment report to provide a scientific basis for subsequent handling of the fire.

[0003] In the process of assessing self-compacting concrete (SC) structural components, accurately determining the fire-affected characteristics of SC is crucial for calculating the residual compressive strength of the structure after a fire and guiding post-disaster repair. However, the existing "Standard for Post-Fire Structural Appraisal of Engineering Structures T / CECS_252—2019" does not include assessment and appraisal calculation parameters related to SC. Therefore, the assessment and appraisal calculation parameters related to ordinary concrete in the same standard are often directly adopted. The problem with this approach is that due to the high density of SC, it faces a more significant risk of performance degradation under fire conditions. Therefore, assessment results based on the assessment and appraisal calculation parameters related to ordinary concrete may lead to insufficient repairs, leaving safety hazards, and even causing secondary structural damage to SC buildings after repair, resulting in greater economic losses.

[0004] If the strength test of standardized specimens of self-compacting concrete after high temperature is used for evaluation, the specimens need to be cured for a specified 28-day period after casting, and then subjected to simulated high temperature treatment and testing. However, this testing method has many disadvantages, such as requiring a large number of standardized specimens, low testing efficiency, high cost, and long time consumption.

[0005] In recent years, with the development of artificial intelligence technology, machine learning has provided a convenient solution for performance assessment and intelligent monitoring of buildings after fires. Machine learning is used to construct predictive models, achieving accurate prediction of the residual compressive strength of self-compacting concrete (SCC) at high temperatures through high-precision fitting of multiple input and output parameters. However, the "black box" nature of traditional machine learning algorithms means that model parameters lack physical meaning, making it difficult to understand the internal working principles of the model. This makes it difficult for engineers to determine the core driving factors and their impact on the loss of residual strength at high temperatures, thus hindering the optimization of relevant mix proportions for self-compacting concrete using predictive models. Furthermore, existing predictive models only cover a few variables such as specimen size and temperature, neglecting the influence of key factors such as fiber type, amount of supplementary cementitious materials, heating rate, and fire exposure time. Therefore, their applicability is narrow and they cannot handle the rapid prediction of residual compressive strength after high-temperature scenarios with diverse SCC mix proportions in actual engineering projects. Summary of the Invention

[0006] To overcome the shortcomings in the prior art, this invention discloses a rapid prediction method and source interpretation system for the high-temperature compressive strength of self-compacting concrete, in order to solve the technical problems faced in the field of post-fire identification of self-compacting concrete engineering structures.

[0007] To achieve the aforementioned objective, the present invention employs the following technical solution: a method for rapid prediction of the high-temperature compressive strength of self-compacting concrete, comprising the following steps:

[0008] S1: Data collected from standard dense concrete specimens and manually reviewed; the dataset includes several types of variable characteristic parameters and high-temperature residual compressive strength characteristic parameters.

[0009] S2: Perform Pearson correlation analysis on the dataset parameters; if the absolute value of the correlation coefficient of two parameters is greater than the set threshold, delete one of the two parameters.

[0010] S3: Based on the machine learning model, the prediction model is trained using the dataset after Pearson correlation analysis, with several class variable feature parameters and high-temperature residual compressive strength feature parameters.

[0011] S4: The completed prediction model uses the SHAP algorithm for source interpretation, analyzes the importance ranking of each variable's characteristic parameters, the conditional expectation of a single feature, and the dependency relationship of multiple features, and finally outputs the results in a visual graphical and numerical manner to explain the influence of a single variable's characteristic parameters and their coupling effects on the characteristic parameters of the high-temperature residual compressive strength of self-compacting concrete.

[0012] S5: Prediction of residual compressive strength after high temperature. By conducting on-site investigation and testing of the building structure after the fire, and based on the relationship between the appearance characteristics of concrete components and temperature given in the "Standard for Appraisal of Engineering Structures after Fire T / CECS_252—2019", the fire-affected characteristic parameters of the building structure are inferred. The mix proportion parameters and fiber parameters are obtained by finding the design and construction data of the building that was burned. The fire-affected characteristic parameters, mix proportion parameters, and fiber parameters are input into the prediction model to predict the residual compressive strength of the self-compacting concrete of the building structure after the fire, and then the residual bearing capacity of the building structure after the fire is calculated.

[0013] Furthermore, the aforementioned variable characteristic parameters include: mix proportion parameters, fiber parameters, and fire-affected characteristic parameters;

[0014] The mix proportion parameters include: cement content, supplementary cementitious material (SCMs) content, water-cement ratio, fine aggregate content, coarse aggregate content, sand ratio, and water-reducing agent content;

[0015] The fiber parameters include fiber type and fiber content; fiber type includes: fiber-free, polypropylene fiber (PP fiber), steel fiber or basalt fiber (BA fiber).

[0016] The fire-affected characteristic parameters include: heating rate, fire-affected temperature, and fire-affected time;

[0017] The characteristic parameters of high-temperature residual compressive strength include: compressive strength.

[0018] Furthermore, based on the individual variable characteristic parameters and their coupling effects output by the prediction model in step S4, the influence of the high-temperature residual compressive strength characteristic parameters of self-compacting concrete on the high-temperature residual compressive strength characteristic parameters is visualized and numerically analyzed to optimize the mix proportion parameters and fiber parameters of self-compacting concrete.

[0019] Furthermore, machine learning models include: Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), or Artificial Neural Network (ANN).

[0020] Furthermore, the training method for machine learning models is as follows:

[0021] S3.1: Save the characteristic parameters of several types of self-compacting concrete and the characteristic parameters of high-temperature residual compressive strength to the database as a dataset; divide the dataset into training set and test set;

[0022] S3.2: The machine learning model is trained using the training set to obtain the corresponding prediction model; the training process uses K-fold cross-validation to improve the generalization ability of the prediction model;

[0023] S3.3: Use the test set to evaluate the performance of the trained prediction model and finally obtain the optimal prediction model.

[0024] Furthermore, during the training of machine learning models, the Sparrow Search Algorithm (SSA) or Whale Optimization Algorithm (WOA) is used to quickly optimize the hyperparameters of the prediction model.

[0025] Furthermore, the performance evaluation of the predictive model includes: the coefficient of determination (R²). 2 The prediction model is evaluated using four indicators: root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). The accuracy and generalization ability of the prediction model are comprehensively evaluated using these four indicators. The formula for calculating the comprehensive evaluation indicator Gen (1) is as follows:

[0026] (1)

[0027] In the formula, the subscripts "tr" and "te" represent the training set and the test set, respectively.

[0028] The coefficient of determination (R) 2 The calculation formulas (2), (3), (4), and (5) for the root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE) are shown below:

[0029] (2);

[0030] (3);

[0031] (4);

[0032] (5);

[0033] Wherein: In the above calculation formula, Yte and Ypre represent the experimental result and the predicted result, respectively; The mean of the test results is represented by N; N represents the number of samples.

[0034] A traceability interpretation system for a rapid prediction method of high-temperature compressive strength of self-compacting concrete includes media for data input, storage, processing, and interactive interface;

[0035] The data input medium is used for data input.

[0036] The storage medium is used to store the dataset and the prediction model.

[0037] Among them, the processing medium is used for training the prediction model and for rapid prediction data processing of the high-temperature residual compressive strength of self-compacting concrete.

[0038] The interactive interface is used for data input and prediction result output.

[0039] Due to the adoption of the technical solution described above, the present invention has the following beneficial effects: The present invention discloses a rapid prediction method and source interpretation system for the high-temperature compressive strength of self-compacting concrete. By collecting mix proportion parameters, fiber parameters, and fire-affected characteristic parameters data in the existing research field on the high-temperature residual compressive strength of self-compacting concrete, a machine learning model is trained to establish an interpretable prediction model and source interpretation system for the high-temperature compressive strength of self-compacting concrete. This enables rapid prediction of the high-temperature compressive strength of self-compacting concrete, providing more accurate, reliable, and scientific intelligent decision support for post-disaster building repair. At the same time, based on the source interpretation of the rapid prediction model for the high-temperature residual compressive strength of self-compacting concrete, engineers can understand the core driving factors and their influence on the loss of high-temperature residual compressive strength of self-compacting concrete by individual variable characteristic parameters and their coupling effects. Thus, the relevant parameters of self-compacting concrete can be optimized through the source interpretation results of the prediction model to improve the high-temperature compressive strength of self-compacting concrete. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the frame structure for predicting the high-temperature compressive strength of self-compacting concrete.

[0041] Figure 2 This is a table showing the statistical results of characteristic parameters of the standard sample dataset for self-compacting concrete.

[0042] Figure 3 A visualization of the calculated Pearson correlation coefficient results for the characteristic parameters of the standard self-compacting concrete sample dataset;

[0043] Figure 4 A visualization of the hyperparameter optimization process during the training of a prediction model for the high-temperature compressive strength of self-compacting concrete.

[0044] Figure 5 Table showing the optimization results of hyperparameters for the prediction model of high-temperature compressive strength of self-compacting concrete;

[0045] Figure 6 This table shows the calculation results of the comprehensive evaluation index of the high-temperature compressive strength prediction model for self-compacting concrete based on the training set and the test set.

[0046] Figure 7 A visualization of the test results of the high-temperature compressive strength prediction model for self-compacting concrete.

[0047] Figure 8 A visualization of the full feature explanation of the SSA-XGBoost prediction model by SHAP;

[0048] Figure 9 A visualization of SHAP's interpretation of a single feature in the SSA-XGBoost prediction model;

[0049] Figure 10 A visualization illustrating the feature interactions of SHAP with the SSA-XGBoost prediction model;

[0050] Figure 11 A visualization of the interpretation of individual sample features by SHAP for the SSA-XGBoost prediction model;

[0051] Figure 12 This is a feature interaction network diagram of SHAP for the SSA-XGBoost prediction model. Detailed Implementation

[0052] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0053] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0054] Some exemplary embodiments of the invention have been described for illustrative purposes. It should be understood that the invention may be implemented in other ways not specifically shown in the accompanying drawings.

[0055] The present invention will be explained in detail through the following embodiments. The purpose of disclosing the present invention is to protect all technical improvements within the scope of the present invention.

[0056] Example 1: A rapid prediction method for the high-temperature compressive strength of self-compacting concrete. Based on machine learning, an interpretable prediction model is established using experimental data from existing related studies. This model is then used to predict the loss of compressive strength of self-compacting concrete after high-temperature exposure, while tracing the causes of this high-temperature strength loss and visualizing the results graphically or in charts. This rapid prediction model for the high-temperature residual compressive strength of self-compacting concrete can be applied to the quantitative calculation of structural assessment of self-compacting concrete buildings after fire, enabling rapid prediction of the high-temperature compressive strength of self-compacting concrete and providing more accurate, reliable, and scientific intelligent decision support for post-disaster building repair. For the specific framework of the high-temperature compressive strength prediction model for self-compacting concrete, please refer to the appendix of the instruction manual. Figure 1 .

[0057] A rapid method for predicting the high-temperature compressive strength of self-compacting concrete includes the following steps:

[0058] S1: Experimental data from existing relevant studies were collected to establish a large dataset of standard self-compacting concrete samples containing 408 test specimens. In this dataset, industrial byproducts, including fly ash, silica fume, and slag, are collectively referred to as supplementary cementitious materials (SCMs). Cementitious materials (CMs) include cement and SCMs. All data in the dataset underwent manual review; samples lacking basic information were excluded to ensure interpretability and reduce potential confounding factors. See the appendix in the instruction manual. Figure 2The table presents the statistical results of the characteristic parameters of the collected specimens: in the table, Min, Max, and Mea represent the minimum, maximum, and average values, respectively; Std represents the standard deviation; and Q25% and Q75% represent the 25th and 75th quantiles, respectively. All input parameters in the database can be directly obtained from the literature. For the output parameters, since the database contains specimen sizes commonly used in different countries and regions, the data was normalized to eliminate the influence of size effects on the output results, converting all specimens to the residual strength corresponding to the standard size of 150mm × 150mm × 150mm. The average range of residual compressive strength after high temperature for all specimens is 41.52 MPa, indicating that the strength range collected in this invention mostly corresponds to ordinary strength concrete. Overall, the established database has a wide range of values ​​and a reasonable parameter distribution, demonstrating significant engineering applicability. It can adapt to advanced machine learning models to predict the residual compressive strength of self-compacting concrete after high-temperature treatment. The dataset includes several classes of variable feature parameters and high-temperature residual compressive strength feature parameters. The variable feature parameters include mix proportion parameters, fiber parameters, and fire-induced characteristic parameters. The high-temperature residual compressive strength feature parameters include compressive strength. The mix proportion parameters include cement content, supplementary gelling material content, water-cement ratio, fine aggregate content, coarse aggregate content, sand ratio, and water-reducing agent dosage. The fiber parameters include fiber type and fiber dosage. Fiber type includes fiberless (input parameter defined as 0), polypropylene fiber (PP fiber, input parameter defined as 1), steel fiber (input parameter defined as 2), or basalt fiber (BA fiber, input parameter defined as 3). The fire-induced characteristic parameters include heating rate, fire temperature, and fire time.

[0059] S2: Perform Pearson correlation analysis on the parameters in the dataset; Pearson correlation analysis reveals the degree of correlation between parameters through the correlation coefficient R. When the absolute value of the correlation coefficient of two parameters is greater than the set threshold of 0.8, it indicates that the two parameters are too linearly correlated. At this time, one of the two parameters can be deleted to reduce the potential impact of multicollinearity; The formula for calculating the correlation coefficient R (6) is shown below;

[0060] (6)

[0061] In the formula, R is the correlation coefficient, representing the degree of linear correlation between the parameters; X i and Y i These represent the i-th row of data for variables X and Y, respectively. and Let X and Y represent the average values ​​of variables X and Y, respectively. The R-squared value ranges from -1 to 1. A value closer to 1 indicates a stronger positive correlation, while a value further away indicates a stronger negative correlation. A value closer to 0 indicates a weaker correlation. Excessive correlation between input variables may reduce the predictive efficiency of the machine learning model. To ensure the effectiveness of the machine learning model, the R-squared value between the two input parameters should be kept below 0.8 to minimize the potential impact of multicollinearity. See the appendix in the instruction manual. Figure 3 The results show the calculated Pearson correlation coefficients for all parameters. The strongest correlation was between sand ratio and coarse aggregate content, with a correlation coefficient of -0.79. A higher sand ratio corresponds to a lower coarse aggregate content; this correlation is determined by the definition of sand ratio, which represents the ratio of fine aggregate content to total aggregate content. Furthermore, fiber type and fiber content showed a significant positive correlation, with a correlation coefficient of 0.77. Additionally, among the investigated characteristics, the firing temperature showed the strongest correlation with the high-temperature residual compressive strength of self-compacting concrete (SCC), with a correlation coefficient of -0.65, indicating that a higher firing temperature results in a lower residual compressive strength for SCC. Firing time and heating rate also showed positive correlations with residual compressive strength. The effects of positive and negative correlations: Among the mix proportion parameters, the fine aggregate content and water-cement ratio have the most significant correlations with the residual compressive strength of SCC after high temperature; specifically, the fine aggregate content has a positive correlation with the target result, while the water-cement ratio has a negative correlation with the predicted result; the correlation between cement and supplementary cementitious materials with the target result is second only to the fine aggregate content and water-cement ratio; specifically, cement has a positive correlation with the residual compressive strength, while supplementary cementitious materials have a negative correlation with the predicted result; this indicates that the high-temperature compressive strength of SCC increases with the increase of cement content, but increases with the decrease of supplementary cementitious material content; the linear correlation between coarse aggregate content, fiber type, fiber content, and water-reducing agent content and the predicted result is not significant;

[0062] S3: Based on a machine learning model, using the dataset after Pearson correlation analysis, a prediction model is trained with several class variable feature parameters and high-temperature residual compressive strength feature parameters; the training method of the prediction model includes the following steps:

[0063] S3.1: Save the characteristic parameters of several types of self-compacting concrete and the characteristic parameters of high-temperature residual compressive strength to the database as a dataset; divide the dataset into a training set and a test set, with 70% of the data assigned to the training set and the remaining 30% assigned to the test set;

[0064] S3.2: The machine learning model is trained using a training set to obtain the corresponding prediction model. The machine learning models include: Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), or Artificial Neural Network (ANN). During the training of these machine learning models, hyperparameter optimization is performed using Sparrow Search Algorithm (SSA) or Whale Optimization Algorithm (WOA). Therefore, a total of six prediction models can be trained. To avoid the prediction model getting trapped in local optima during the machine learning model training process, K-fold cross-validation is used to optimize the hyperparameters of the prediction model, where K is typically set between 5 and 10. In this embodiment, the training set is randomly divided into five training subsets of equal (or approximately equal) size. In each round, one training subset is used as the test set, and the remaining four training subsets are used as the training set. The machine learning model is trained on the training sets, and finally, the performance of the trained prediction model is evaluated on the test set. The aforementioned Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), Artificial Neural Network (ANN), Sparrow Search Algorithm (SSA), and Whale Optimization Algorithm (WOA) are all existing technologies and will not be described in detail here. For the hyperparameter optimization process during machine learning model training, please refer to the appendix of the specification. Figure 4 For hyperparameter optimization results, please refer to the appendix of the instruction manual. Figure 5 ;

[0065] S3.3: The trained prediction models are evaluated using a test set. The test results for the six final trained prediction models are attached to the instruction manual. Figure 6 R-values ​​of the six prediction models on the training and test sets. 2 All values ​​are greater than 0.9, indicating that all prediction models have high accuracy in predicting the output results, with the SSA-XGBoost model showing the highest R-value. 2 The value was highest on both the training set (0.987) and the test set (0.948); to further comprehensively evaluate the performance of the prediction model, the performance evaluation was based on the output results of the training and test sets, using the coefficient of determination (R²). 2 The accuracy and generalization ability of the prediction model are comprehensively evaluated using four indicators: root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). The specific comprehensive evaluation calculation formula (1) is shown below:

[0066] (1)

[0067] Where Gen represents the comprehensive evaluation index of the prediction model, and the smaller the calculated value, the better; where the subscripts “tr” and “te” in formula (1) represent the training set and the test set, respectively, the calculation results of the comprehensive evaluation index of the prediction model can be found in the appendix of the instruction manual. Figure 6Among the six prediction models, the SSA-XGBoost model showed the smallest Gen value (8.285), indicating that SSA-XGBoost has excellent prediction accuracy and the smallest prediction error. Finally, SSA-XGBoost was selected as the optimal prediction model for rapid prediction of high temperature compressive strength of self-compacting concrete.

[0068] S4: For the completed SSA-XGBoost optimal prediction model, the SHAP algorithm is used to analyze the importance of each variable's characteristic parameters, the conditional expectation of a single feature, and the dependency relationship of multiple features. Finally, the results are output in a visual and numerical manner. Through the interpretation of all feature rankings, the interpretation of the influence of a single feature, the interpretation of feature interactions, and the interpretation of single sample features, the influence of the characteristic parameters of a single variable and their coupling effects on the characteristic parameters of the high-temperature residual compressive strength of self-compacting concrete is clarified, thus providing in-depth guidance for revealing the causes of high-temperature strength loss in self-compacting concrete and its improvement.

[0069] S5: Prediction of residual compressive strength after high temperature. On-site investigation and testing of the building structure after the fire are conducted. Based on the relationship between the appearance characteristics of concrete components and temperature given in the "Standard for Appraisal of Engineering Structures After Fire T / CECS_252—2019", the fire-affected characteristic parameters of the building structure are inferred (see "Standard for Appraisal of Engineering Structures After Fire T / CECS_252—2019" for specific methods). Mix proportion parameters and fiber parameters are obtained from the design and construction data of the fire-affected building. These parameters are then input into the prediction model. Through visualization and numerical interpretation of all features, the residual compressive strength of the self-compacting concrete at any depth point of the fire-affected building structure is predicted. This predicted residual compressive strength at any depth point is more accurate than the compressive strength reduction factor for concrete after natural cooling at high temperature in Appendix Table G.0.1-1 of the existing "Standard for Appraisal of Engineering Structures After Fire T / CECS_252—2019", thus resulting in a more accurate calculation of the residual load-bearing capacity of the building structure after the fire.

[0070] Example 2: In this example, the source tracing explanation in step S4 is specifically elaborated: see the appendix to the specification. Figure 8 This shows the explanation of SHAP's ranking of the importance of all features in the optimal prediction model of SSA-XGBoost: Appendix Figure 8The left side shows the SHAP feature importance of all input parameters, where the global importance of each feature is considered as the average absolute value of that feature across all given samples. The x-axis represents the average absolute value of the SHAP values ​​of each input parameter, and the y-axis represents the importance ranking of each influencing factor. As shown in the figure, among the fire-affected feature parameters, fire temperature has an absolute advantage in contributing to the prediction results, with an average absolute value of 0.6559. In contrast, the average absolute values ​​of the SHAP values ​​of fire time and heating rate are only 0.0273 and 0.0115, respectively, contributing very little to the prediction results. Secondly, concrete mix proportion parameters have a significant impact on the prediction results. The cumulative contribution of all concrete mix proportion parameters to the prediction results is slightly greater than that of fire temperature, as the sum of their average absolute values ​​of SHAP is 0.8671. Among all mix proportion parameters, fine aggregate content, water-cement ratio, and supplementary cementitious materials have the best contributions. Fiber type contributes slightly more to the prediction results than fiber content.

[0071] Appendix Figure 8The right side displays the SHAP honeycomb diagrams for each feature, providing the degree of significance of each input parameter's influence on the prediction results for each sample. The x-axis represents the magnitude of the Shapley value; darker yellow scatter points indicate a significant negative impact of the input parameter on the prediction target, while darker blue scatter points indicate a significant positive impact. The y-axis represents the ranking of model parameters by importance; feature parameters farther from the y-axis baseline have a more significant contribution to the prediction results. As shown in the figure, the fire temperature contributes most significantly to the prediction results. Higher fire temperatures result in smaller Shaple values ​​for the compressive strength of SCC concrete after fire, indicating that the residual compressive strength has a greater negative impact with increasing temperature. Higher fine aggregate content leads to larger Shaple values, indicating that in SCC concrete after high temperatures, increasing the fine aggregate content has a positive effect on compressive strength. This is because increasing the fine aggregate content can improve the microstructure of the concrete. Density inhibits the shrinkage and cracking of the paste at high temperatures, thus improving the residual compressive strength of SCC concrete after high-temperature treatment. Furthermore, changes in the water-cement ratio significantly affect the output results; a lower water-cement ratio results in a higher Shap value, indicating that the residual compressive strength increases with decreasing water-cement ratio. This effect is consistent with the experience gained from compressive strength tests on SCC concrete at room temperature. Regarding the content of supplementary cementitious materials and cement, their contributions to the residual compressive strength of SCC concrete after high-temperature treatment are opposite. Higher amounts of supplementary cementitious materials result in lower residual compressive strength, while higher amounts of cement result in higher residual compressive strength. This is mainly because supplementary cementitious materials have lower reactivity, leading to hydration products that are less effective than cement. In addition, some other characteristics, including sand ratio, water-reducing agent, firing time, and fiber type, negatively impact the residual strength of concrete after high-temperature treatment as their characteristic values ​​increase, although the degree of influence of these parameters is relatively small.

[0072] See the instruction manual appendix Figure 9 This demonstrates the explanation of the impact of individual features, clearly showing how the prediction results of a single sample change when the six most important input features change independently, thus visualizing the relationship between a single feature variable and the target result in each sample; (In the appendix...) Figure 9As shown in (a), the firing temperature has a negative impact on the residual compressive strength of SCC concrete after high-temperature treatment: with the increase of the firing temperature, the residual compressive strength of SCC concrete after high-temperature treatment will drop sharply twice. When the firing temperature reaches about 450℃, the residual compressive strength will drop sharply by about 48.40%; when the firing temperature reaches about 800℃, the residual compressive strength will continue to drop sharply by about 82.84%. The first sharp drop in compressive strength is due to the high temperature forcing calcium hydroxide to decompose, resulting in a significant decrease in the cementitious bonding force, causing the cracks in the interface transition zone to penetrate, and the aggregate and paste to fail to bond, thereby reducing the compressive strength. The second sharp drop in compressive strength is due to the high temperature forcing the complete decomposition of calcium hydroxide, generating thermal stress between the aggregate and paste, causing the cracks to penetrate, and at the same time, the internal pore pressure accumulates to exceed the tensile strength, ultimately leading to the loosening of the internal structure, thereby reducing the compressive strength again; as shown in the attached figure. Figure 9 As shown in (b), when the fine aggregate content is 800~900 kg / m³, increasing the fine aggregate content has the most significant effect on the residual compressive strength; from the attached... Figure 9 (c) It can be seen that the residual compressive strength of SCC concrete after high temperature decreases with the increase of water-cement ratio. When the water-cement ratio reaches 0.55, increasing the water-cement ratio has almost no effect on the residual compressive strength. In addition, the amount of cementitious materials and cement admixtures has little effect on the residual compressive strength of SCC concrete after high temperature, as shown in the attached figure. Figure 9 As shown in (d) and (e); from Figure 9 (f) It can be seen that when the sand ratio is maintained at about 0.484, the residual compressive strength of SCC concrete after high temperature is relatively high. After that, if the sand ratio is further increased, the residual compressive strength will decrease slightly.

[0073] See the instruction manual appendix Figure 10 This demonstrates the interpretation of feature interactions, and the visualization shows the extent to which significant interactions between local features affect the prediction results; Appendix Figure 10 (a) The rapid loss of high-temperature residual compressive strength can be attributed to the interaction between firing temperature and fine aggregate content; Appendix Figure 10 (b) shows the effect of the coupling effect of fire temperature and water-cement ratio on compressive strength. As shown in the figure, at room temperature, the water-cement ratio is strongly negatively correlated with compressive strength; that is, the smaller the water-cement ratio, the higher the compressive strength, and the larger the water-cement ratio, the lower the compressive strength. However, at high temperatures (such as 600℃), the effect of the water-cement ratio shows a moderate range. When the water-cement ratio is around 0.4, the residual strength of the concrete is relatively higher. Too small a water-cement ratio can easily lead to vapor pressure bursting due to excessively dense cement matrix, while too large a water-cement ratio can easily lead to structural compressive strength loss due to excessive initial porosity. The interaction between fire temperature and supplementary cementitious materials on compressive strength is shown in the appendix. Figure 10As shown in (c), at room temperature, the replacement of some cement with supplementary cementitious materials leads to a slight decrease in compressive strength; at high temperatures, the addition of supplementary cementitious materials results in insufficient hydration products, leading to a loose structure and thus accelerating the loss of compressive strength; Appendix Figure 10 (d) shows the effect of the coupling effect of fire temperature and cement on compressive strength. The observed trend shows that at room temperature, the higher the cement content, the more cementitious materials participate in hydration, and the bonding strength and density of the cement matrix are significantly improved, thereby increasing the compressive strength. At high temperature, the high cement content promotes the hydration reaction, resulting in some cementitious products that can delay the collapse of the overall structure. At the same time, the dense matrix makes it difficult for high temperature steam to accumulate quickly to form high pressure, reducing the risk of high temperature cracking, thereby effectively reducing the propagation rate of microcracks caused by thermal stress and indirectly improving the residual strength after high temperature.

[0074] See the instruction manual appendix Figure 11 The figure vividly illustrates the influence of individual sample features, using a waterfall plot of feature interpretations from two representative sets of samples to reveal the underlying reasons why each input feature affects the prediction result of a single sample. As shown in the figure, the baseline expectation reflects the average prediction result of the target parameter in the entire dataset. In this model, the baseline expectation value is set to 42.401. The orange and blue bars indicate whether the contribution of each input parameter relative to the baseline pushes up or down the prediction result, respectively. The length of the bar indicates the magnitude of the increase or decrease in response.

[0075] The first group describes the interpretation results for two specimens with only varying firing temperatures. The firing temperatures for the two specimens were 400℃ and 600℃, respectively. As observed, the residual compressive strength of both specimens after high-temperature firing should initially be near the baseline value. However, the different firing temperatures caused the actual predicted results for the two specimens to tend in opposite directions. A firing temperature of 400℃ helped increase the predicted result by 5.03 MPa, while a firing temperature of 600℃ caused the predicted result to decrease by 16.27 MPa. The explanation for this result is that at 400℃, Ca(OH)₂ undergoes initial dehydration, producing finer CaO particles that can fill internal pores. Additionally, as water is gradually released, the CSH gel structure in the cement paste shrinks, slightly increasing the matrix density and stress. The uniform heat transfer slightly slowed the decline in residual strength; however, at 600℃, Ca(OH)2 was completely dehydrated, generating a large amount of CaO, leading to dense microcracks inside the matrix. Furthermore, the CSH in the cement paste decomposed into CaO and SiO2, destroying the cementitious structural skeleton and drastically reducing the bond strength between the cement paste and aggregate, resulting in rapid strength degradation. Based on these reasons, considering the combined effects of all characteristic parameters, the predicted strength for the sample at 400℃ was improved from the baseline value to 43.987 MPa, showing good agreement with the measured value of 43.74 MPa. Conversely, the predicted strength for the sample at 600℃ was reduced from the baseline value to 19.422 MPa, which is in good agreement with the measured value of 18.8 MPa.

[0076] To further reveal the mechanism of feature coupling in complex models, a feature interaction network graph is used in step S4 of the prediction model to elucidate the impact of feature coupling, thereby achieving a more comprehensive understanding of the prediction results. (See appendix to the instruction manual.) Figure 12 The graph shows the feature interaction network based on SHAP for each parameter. The graph illustrates the importance of the feature parameters and their interactions through nodes and edges. The size and color of the nodes indicate the contribution of the feature to the output, while the thickness and color of the edges represent the interaction effects between features.

[0077] As attached Figure 12As shown, firing temperature, fine aggregate content, and water-cement ratio are the three features that contribute most to the output results. Among them, the interaction effect between firing temperature and water-cement ratio is the most significant, with an interaction value of -0.125, indicating that the coupling effect between the two has the highest degree of influence on the target result. In addition, the two show a negative interaction contribution, indicating that when the firing temperature and the water-cement ratio increase, it will have a more significant negative impact on the residual compressive strength after high temperature. This is because SCC with a high water-cement ratio has more pores, and is more prone to cracking due to moisture evaporation and thermal expansion after firing. The increase in firing temperature will aggravate this damage. When both are at high levels, the performance degradation of the material after firing is more significant; the interaction effect between firing temperature and fine aggregate content is also quite obvious, with an interaction value of -0.104, and the two also show a negative interaction contribution; this is basically consistent with the interaction explanation of firing temperature and water-cement ratio; it is worth mentioning that the supplementary cementitious content has a positive interaction contribution with the firing temperature, which indicates that as the firing temperature increases, increasing the amount of supplementary cementitious material will alleviate the degradation of compressive strength of SCC after high temperature to a certain extent; therefore, the addition of supplementary cementitious material helps to improve the high-temperature compressive performance of SCC;

[0078] Based on the source interpretation of the above-mentioned rapid prediction model for the high-temperature residual compressive strength of self-compacting concrete, engineers can understand the core driving factors and their degree of influence of individual variable characteristic parameters and their coupling effects on the loss of high-temperature residual compressive strength of self-compacting concrete. Thus, the relevant mix proportion parameters of self-compacting concrete can be optimized through the source interpretation results of the prediction model, thereby improving the high-temperature compressive strength of self-compacting concrete.

[0079] A source-tracing and interpretation system for a rapid prediction method of high-temperature compressive strength of self-compacting concrete includes media for data input, storage, processing, and an interactive interface. The data input medium is used for data input; the storage medium is used for storing datasets and prediction models; the processing medium is used for training the prediction model and processing data for rapid prediction of the mechanical characteristics of self-compacting concrete after high temperatures; and the interactive interface is used for interaction between data input and prediction result output. Specifically, the source-tracing and interpretation system for the rapid prediction method of high-temperature compressive strength of self-compacting concrete can be a computer system storing datasets and prediction models, or a computer network based on datasets and prediction models stored in a computing center. By inputting fire-affected characteristic parameters, mix proportion parameters, and fiber parameters of a self-compacting concrete building that has experienced a fire, the system automatically calculates and predicts the residual compressive strength after the fire. Through visual graphics and specific numerical values, it displays and explains the root causes of the high-temperature strength loss of self-compacting concrete and provides objective data for calculating the residual load-bearing capacity of buildings after a fire, thereby providing more objective, accurate, reliable, and scientific intelligent decision support for post-disaster building repair.

[0080] It should be understood that this solution is not limited to the specific embodiments described above. Devices and structures not described in detail herein should be understood as being implemented in a manner common to the art. Any person skilled in the art can make many possible variations and modifications to this solution, or modify it into equivalent embodiments, without departing from the scope of this solution, using the methods and techniques disclosed above. This does not affect the substantive content of this solution. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this solution, without departing from its scope, still fall within the protection scope of this solution.

[0081] The parts of this invention not described in detail are prior art.

Claims

1. A method for rapid prediction of the high-temperature compressive strength of self-compacting concrete, characterized by: Includes the following steps: S1: Data collected from standard dense concrete specimens and manually reviewed; the dataset includes several types of variable characteristic parameters and high-temperature residual compressive strength characteristic parameters. S2: Perform Pearson correlation analysis on the dataset parameters; if the absolute value of the correlation coefficient of two parameters is greater than the set threshold, delete one of the two parameters. S3: Based on the machine learning model, the prediction model is trained using the dataset after Pearson correlation analysis, with several class variable feature parameters and high-temperature residual compressive strength feature parameters. S4: The completed prediction model uses the SHAP algorithm for source interpretation, analyzes the importance ranking of each variable's characteristic parameters, the conditional expectation of a single feature, and the dependency relationship of multiple features, and finally outputs the results in a visual graphical and numerical manner to explain the influence of a single variable's characteristic parameters and their coupling effects on the characteristic parameters of the high-temperature residual compressive strength of self-compacting concrete. S5: Prediction of residual compressive strength after high temperature. By conducting on-site investigation and testing of the building structure after the fire, and based on the relationship between the appearance characteristics of concrete components and temperature given in the "Standard for Appraisal of Engineering Structures after Fire T / CECS_252—2019", the fire-affected characteristic parameters of the building structure are inferred. The mix proportion parameters and fiber parameters are obtained by finding the design and construction data of the building that was burned. The fire-affected characteristic parameters, mix proportion parameters, and fiber parameters are input into the prediction model to predict the residual compressive strength of the self-compacting concrete of the building structure after the fire, and then assess the residual load-bearing capacity of the building structure after the fire.

2. The method for rapid prediction of high-temperature compressive strength of self-compacting concrete according to claim 1, characterized in that: The aforementioned variable characteristic parameters include: mix proportion parameters, fiber parameters, and fire-affected characteristic parameters; The mix proportion parameters include: cement content, supplementary cementitious material (SCMs) content, water-cement ratio, fine aggregate content, coarse aggregate content, sand ratio, and water-reducing agent content; The fiber parameters include fiber type and fiber content; fiber type includes: fiber-free, polypropylene fiber (PP fiber), steel fiber or basalt fiber (BA fiber). The fire-affected characteristic parameters include: heating rate, fire-affected temperature, and fire-affected time; The characteristic parameters of high-temperature residual compressive strength include: compressive strength.

3. The method for rapid prediction of high-temperature compressive strength of self-compacting concrete according to claim 1, characterized in that: Based on the individual variable characteristic parameters and their coupling effects output by the prediction model in step S4, the visualization graphs and numerical values ​​of their influence on the characteristic parameters of the high-temperature residual compressive strength of self-compacting concrete are used to optimize the mix proportion parameters and fiber parameters of self-compacting concrete.

4. The method for rapid prediction of high-temperature compressive strength of self-compacting concrete according to claim 1, characterized in that: Machine learning models include: Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), or Artificial Neural Network (ANN).

5. The method for rapid prediction of high-temperature compressive strength of self-compacting concrete according to claim 1, characterized in that: The training method for machine learning models is as follows: S3.1: Save the characteristic parameters of several types of self-compacting concrete and the characteristic parameters of high-temperature residual compressive strength to the database as a dataset; divide the dataset into training set and test set; S3.2: The machine learning model is trained using the training set to obtain the corresponding prediction model; the training process uses K-fold cross-validation to improve the generalization ability of the prediction model; S3.3: Use the test set to evaluate the performance of the trained prediction model and finally obtain the optimal prediction model.

6. The method for rapid prediction of high-temperature compressive strength of self-compacting concrete according to claim 5, characterized in that: During the training of machine learning models, the Sparrow Search Algorithm (SSA) or Whale Optimization Algorithm (WOA) are used to quickly optimize the hyperparameters of the prediction model.

7. The method for rapid prediction of high-temperature compressive strength of self-compacting concrete according to claim 6, characterized in that: Performance evaluation of predictive models includes: coefficient of determination (R²) 2 The prediction model is evaluated using four indicators: root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE). The accuracy and generalization ability of the prediction model are comprehensively evaluated using these four indicators. The formula for calculating the comprehensive evaluation indicator Gen (1) is as follows: （1） In the formula, the subscripts "tr" and "te" represent the training set and the test set, respectively. The coefficient of determination (R) 2 The calculation formulas (2), (3), (4), and (5) for the root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE) are shown below: （2）； （3）； （4）； （5）； Where: In the above calculation formula, Y te and Y pre These represent the experimental results and the predicted results, respectively. The mean of the test results is represented by N; N represents the number of samples.

8. A source interpretation system for the rapid prediction method of high-temperature compressive strength of self-compacting concrete according to any one of claims 1 to 7, characterized in that: This includes media for data input, storage, processing, and interactive interfaces; The data input medium is used for data input. The storage medium is used to store the dataset and the prediction model. Among them, the processing medium is used for training the prediction model and for rapid prediction data processing of the high-temperature residual compressive strength of self-compacting concrete. The interactive interface is used for data input and prediction result output.

Images