A method for predicting compressive strength of alkali-activated slag material based on thermodynamic simulation

Abstract

The application provides a kind of alkali slag material compressive strength prediction method based on thermodynamic simulation, belongs to building material intelligent design and data driven prediction technical field;The application provides to collect the original data of alkali slag material system, obtains the thermodynamic characteristic parameter capable of representing the internal reaction state of material based on thermodynamic simulation;The basic mix proportion parameter is fused with the thermodynamic characteristic parameter, and the machine learning training data set is constructed;Machine learning algorithm is used to train the training data set, and the alkali slag material compressive strength prediction model is established;The basic mix proportion parameter of the alkali slag material system to be predicted and the corresponding thermodynamic characteristic parameter are input into the trained prediction model, and the compressive strength prediction result under the corresponding age is output;The application realizes high-precision prediction of the compressive strength of alkali slag material system by combining thermodynamic simulation results with machine learning method, improves the model prediction accuracy and generalization ability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent design and data-driven prediction technology of building materials, and relates to a method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation. Background Technology

[0002] Industrial solid wastes such as blast furnace slag are produced in huge quantities. If they are not effectively utilized, they will not only waste resources but may also cause environmental pollution. Alkali blast furnace slag cementitious materials are a new type of low-carbon cementitious material that uses granulated blast furnace slag as the main raw material and forms cementitious properties under the action of alkaline activators. They have the characteristics of low carbon dioxide emissions, high early strength, excellent durability and high solid waste utilization rate, and have broad application prospects in the fields of green building materials and low-carbon cementitious materials.

[0003] However, the mechanical properties of alkali slag materials are influenced by a variety of factors, including the slag chemical composition, activator type, activator modulus, Na₂O equivalent, water-cement ratio, curing temperature, and curing age. These factors exhibit complex nonlinear coupling relationships, and even small changes in different factors can lead to significant fluctuations in material strength. Therefore, the mix design and performance optimization of alkali slag materials currently rely heavily on extensive experimentation, resulting in long testing cycles, high material consumption, and difficulties in achieving rapid and efficient performance prediction and material design.

[0004] With the development of artificial intelligence technology, machine learning methods are increasingly being applied to the prediction of concrete and cementitious material properties. For example, algorithms such as random forests, support vector machines, artificial neural networks, and gradient boosting trees are used to establish a mapping relationship between material mix proportions and compressive strength, thereby enabling material property prediction. Compared to traditional empirical formulas, machine learning models have certain advantages in fitting complex nonlinear relationships. However, most current prediction models rely primarily on macroscopic mix proportion parameters and experimental data for training, lacking consideration of the internal reaction mechanisms and hydration product evolution processes of alkali-slag material systems. This leads to problems such as insufficient generalization ability, weak interpretability, and limited adaptability to different systems. Summary of the Invention

[0005] The technical solution of this invention is used to solve the problem of how to improve the accuracy of compressive strength prediction for alkali slag material systems.

[0006] The present invention solves the above-mentioned technical problems through the following technical solutions: This invention provides a method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation, comprising the following steps: Step 1: Collect raw data of the alkali slag material system; Step 2: Simulate and calculate the hydration reaction process of the alkaline slag material system under different curing ages based on thermodynamic simulation methods to obtain thermodynamic characteristic parameters that can characterize the internal reaction state of the material. The thermodynamic characteristic parameters are calculated as follows: The Cemdata18 thermodynamic database, customized for cementitious material systems, is used. Based on the principle of minimum Gibbs free energy, an equilibrium criterion for minimizing Gibbs free energy in the alkali slag material system is established. The MPK kinetic model was adopted, combining three rate control mechanisms: nucleation and growth, diffusion, and shell chemical reaction. The minimum value among the three rates was taken as the total reaction rate of the alkali slag material system. The degree of hydration of slag at different curing ages was obtained by integrating the total reaction rate over time. The thermodynamic simulation temperature is set, and the actual amount of slag involved in the reaction at each age is used as the boundary condition. The data is then input into the thermodynamic simulation software for quantitative calculation of phase equilibrium. The stacking evolution diagram of hydration products at different ages is output, and thermodynamic characteristic parameters are extracted from the evolution diagram.

[0007] Step 3: Fuse the basic mix proportion parameters and thermodynamic characteristic parameters in the original data, and construct a machine learning training dataset after data preprocessing; Step 4: Use machine learning algorithms to train the training dataset and establish a predictive model for the compressive strength of alkali slag materials; Step 5: Input the basic mix proportion parameters and corresponding thermodynamic characteristic parameters of the alkali slag material system to be predicted into the trained prediction model, and output the predicted compressive strength at the corresponding age.

[0008] Furthermore, the formula for the balance criterion is as follows:

[0009] in, The total Gibbs free energy of the alkali slag material system; The total number of chemical components in the alkali slag material system; For the first The number of moles of each component; For the first The chemical potential of a component in its current state.

[0010] Furthermore, the formula for calculating the total reaction rate is as follows:

[0011] in, The total reaction rate is... The degree of slag hydration, The reaction rate during the nucleation and growth control phases. To control the reaction rate during the diffusion phase, To control the reaction rate in the shell chemical reaction stage, For time.

[0012] Furthermore, the formula for calculating the reaction rate of the nucleation and growth control stage is as follows:

[0013] The formula for calculating the reaction rate in the diffusion-controlled stage is as follows:

[0014] in, This represents the reaction rate constant during the nucleation and growth control phases. The reaction rate constant during the diffusion-controlled phase, This is the shape index.

[0015] Furthermore, the formula for calculating the reaction rate in the shell chemical reaction control stage is as follows:

[0016] in, The rate constant for the shell chemical reaction controlling the reaction phase. This is the mechanism index.

[0017] Furthermore, the thermodynamic simulation method employs one of the following software suites: GEMS, PHREEQC, or FactSag.

[0018] Furthermore, the thermodynamic characteristic parameters include: C-(N)-ASH gel content, porosity, hydrotalcite-like phase content, and OH content. - Concentration, concentration of some alkali metal ions, proportion of unreacted slag, and content of AFt and AFm phases.

[0019] Furthermore, the basic mix proportion parameters include: slag chemical composition, activator type, activator modulus, alkali equivalent, and water-cement ratio.

[0020] Furthermore, the machine learning algorithm employs one of the following: random forest algorithm, support vector machine algorithm, artificial neural network algorithm, and XGBoost algorithm.

[0021] Furthermore, the learning rate of the XGBoost algorithm is set to 0.05, the maximum tree depth is set to 6, and the subsampling rate is set to 0.8.

[0022] The beneficial effects of this invention are as follows: This invention first collects experimental data from different alkali slag material systems to establish a raw database containing information such as slag chemical composition, activator parameters, water-cement ratio, curing conditions, and compressive strength of different near-term samples. This database can comprehensively reflect the influence of different factors on the strength development of alkali slag material systems, providing basic data support for subsequent model training.

[0023] This invention, based on the raw material composition and proportioning parameters of an alkali slag material system, uses thermodynamic simulation to simulate the internal hydration reaction process of the system, thereby obtaining the C-(N)-ASH gel content, porosity, hydrotalcite-like phase content, and OH content. - The thermodynamic characteristic parameters include the concentration of certain alkali metal ions, the proportion of unreacted slag, and the content of AFt and AFm-like phases. Since these thermodynamic parameters can reflect the evolution of phase composition within the alkali slag material system and the changes in ion concentration in the pore solution, compared with the traditional prediction method that only uses macroscopic mix proportion parameters, this invention enables the model to simultaneously consider the macroscopic mix proportion information of the material and the internal chemical reaction mechanism, thereby reducing the prediction error caused by fluctuations in raw materials and differences in the system.

[0024] This invention integrates thermodynamic characteristic parameters with original mix proportion parameters to construct a machine learning training dataset that simultaneously contains macroscopic parameters and microscopic reaction information. Machine learning algorithms are then used to learn the complex nonlinear relationships between different parameters, thereby achieving accurate prediction of the compressive strength of alkali slag material systems. Furthermore, the introduction of thermodynamic simulation parameters enables the model to reflect the internal hydration reaction process and phase composition evolution of the material to a certain extent, thus improving the model's adaptability and mechanism explanation capabilities for different alkali-activated systems.

[0025] This invention allows for the rapid output of compressive strength prediction results at corresponding ages by inputting the mix proportion parameters and thermodynamic characteristic parameters of the system to be predicted into a trained prediction model. This reduces the time and material consumption caused by a large number of experiments in the traditional mix proportion design process, improves the material design efficiency and performance optimization efficiency of alkali slag material systems, and provides a new technical means for the rapid design and engineering application of alkali-activated cementitious material systems. Attached Figure Description

[0026] Figure 1 This is a flowchart of the compressive strength prediction method for alkali slag materials based on thermodynamic simulation according to an embodiment of the present invention; Figure 2 This is a product phase diagram showing the change in the phase equilibrium of the solid product and the porous solution with the proportion of reactant input in an embodiment of the present invention. Figure 3 This is a diagram showing the stacking evolution of hydration products at different curing ages, calculated using the MPK kinetic model in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments: Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are as follows: Step (1): Collect experimental data of the alkali slag material system and establish an original database; based on the basic mix proportion parameters in the original database, use thermodynamic simulation software to simulate the hydration reaction of the alkali slag material system and calculate the thermodynamic characteristic parameters.

[0029] The original database contains 216 sets of experimental data. Each set of experimental data includes parameters such as slag chemical composition, activator type, activator modulus, alkali equivalent, water-cement ratio, curing temperature, curing age, and compressive strength.

[0030] The main chemical composition of the slag includes: CaO 36.2 wt.%, SiO2 34.7 wt.%, Al2O3 14.5 wt.%, and MgO 8.6 wt.%; the activators include at least one of NaOH, KOH, Na2CO3, Na2SO4, water glass, and composite activators.

[0031] Experimental data were standardized and outliers were removed to improve data quality and model training stability.

[0032] The basic mix proportion parameters include: slag chemical composition, activator type, activator modulus, alkali equivalent, and water-cement ratio.

[0033] The thermodynamic simulation software can be one of the following: GEMS, PHREEQC, or FactSag series software.

[0034] The thermodynamic characteristic parameters include: C-(N)-ASH gel content, porosity, hydrotalcite phase content, and OH content. - Concentration, concentration of some alkali metal ions, proportion of unreacted slag, and content of AFt and AFm phases.

[0035] Step (2): Extract the C-(N)-ASH gel content as a thermodynamic characteristic parameter, such as... Figure 2 and Figure 3 The light green area is shown. Since C-(N)-ASH gel is the main cementing product in the alkali slag material system, it can form a continuous gel skeleton structure within the system and plays a major role in the development of the material's compressive strength. Therefore, its content variation can effectively reflect the degree of formation of the effective load-bearing structure within the system. Compared to using only macroscopic mix proportion parameters, using C-(N)-ASH gel content as a model input feature allows the machine learning model to further establish the correlation between the evolution of hydration products and compressive strength, thereby improving the model's prediction accuracy and mechanistic explanation capability.

[0036] The steps for calculating the above thermodynamic characteristic parameters are as follows: The Cemdata18 thermodynamic database, specifically designed for cementitious material systems, was selected. This database integrates the standard thermodynamic properties (such as standard molar Gibbs free energy) of the main hydrated phases in alkali slag material systems (including C-(N)-ASH gel, hydrotalcite-like phases, AFt and AFm-like phases, etc.) and various ions in the pore solution. Standard molar enthalpy The equilibrium state of the reaction system is solved based on the principle of minimum Gibbs free energy, and its core thermodynamic equilibrium criterion formula is shown in equation (1): (1) in, The total Gibbs free energy of the system; The total number of possible chemical components in the system; For the first The number of moles of each component; For the first The chemical potential of a component in its current state.

[0037] The thermodynamic simulation ages include 1d, 3d, 7d, and 28d, and the simulation temperature is set to 25℃. By performing nonlinear programming on equation (1) under the constraint of mass conservation, the output can be obtained as shown in the attached figure. Figure 2 The diagram shows the product phase diagram of the solid product and pore solution phase equilibrium as the reactant input ratio changes. To introduce a time scale and simulate the dynamic evolution at different curing ages, this embodiment further employs the MPK (Modified Parrot and Killoh) kinetic model to calculate the degree of hydration of the slag at different ages. The MPK kinetic model comprehensively considers three rate-controlling mechanisms: nucleation and growth (NG), diffusion (D), and shell chemical reaction (H). The reaction rates under each mechanism are shown in equations (2), (3), and (4), respectively. Reaction rates during nucleation and growth control stages: (2) Reaction rate in the diffusion-controlled phase: (3) The reaction rate in the shell chemical reaction-controlled phase is: (4) in, The degree of slag hydration; , , These are the reaction rate constants for each stage; , The shape and mechanism index. In actual hydration simulations, the overall reaction rate of the alkali slag material system... Following the "barrel effect," we take the minimum value among the three, that is: (5) By time By performing integral calculations, the degree of slag hydration in the alkali-slag material system at specific ages (1d, 3d, 7d, and 28d) can be obtained. The actual amount of slag involved in the reaction at each age is then input as boundary conditions into thermodynamic simulation software for quantitative phase equilibrium calculations. The final output is shown in the attached figure. Figure 3 The diagram shows the stacking evolution of hydration products at different ages. From this evolution data, the content of C-(N)-ASH gel, porosity, hydrotalcite-like phase content, and OH groups were extracted. - Quantitative indicators such as concentration, concentration of some alkali metal ions, proportion of unreacted slag, and content of AFt and AFm phases are used as input feature parameters for machine learning models.

[0038] Step (3): Integrate the basic mixing ratio parameters in the original database with the C-(N)-ASH gel content to establish a training dataset. Preprocess the machine learning training dataset, including at least one of the following: data standardization, missing value handling, outlier removal, and dataset partitioning, to ensure the integrity of the input feature vectors. Divide the machine learning training dataset into a training set and a test set, with the training set accounting for 70-90% of the total dataset.

[0039] Step (4): A predictive model for the compressive strength of alkali slag material is established using the random forest (RF) algorithm. The training set and the test set are divided in an 8:2 ratio. Cross-validation and hyperparameter optimization methods are used to improve the model's prediction accuracy.

[0040] Step (5): Input the test set data into the trained RF algorithm prediction model and output the prediction results of the compressive strength at the corresponding age.

[0041] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 It is 0.84.

[0042] Traditional machine learning-based strength prediction methods rely solely on empirical mappings between mix proportion parameters and experimental strength data. Essentially, these are macroscopic statistical fitting models that lack consideration for the internal hydration reaction mechanisms and phase composition evolution of the material. Consequently, their interpretability is weak, and they are prone to significant prediction biases under conditions of varying curing temperatures, extreme water-cement ratios, or novel systems not found in databases. This invention calculates the internal hydration reaction process of alkali slag materials through thermodynamic simulation. It uses thermodynamic characteristic parameters that characterize the internal reaction state of the material, such as C-(N)-ASH gel, hydrotalcite-like phases, AFt and AFm-like phases, pore solution ion concentration, and the proportion of unreacted slag, as input features to the machine learning model. Since these thermodynamic parameters comprehensively reflect the influence of raw material composition, water-cement ratio, alkali equivalent, curing temperature, and age on the hydration reaction and phase composition evolution of the system, this invention achieves a shift from traditional empirical "mix proportion-strength" fitting to mechanism-based prediction of "mix proportion-hydration reaction-phase composition-strength," thereby improving the model's prediction accuracy and its ability to explain the strength formation mechanism.

[0043] Example 2 This embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Embodiment 1, except that: In step (2), in addition to extracting the C-(N)-ASH gel content, porosity is further extracted as a thermodynamic characteristic parameter. Porosity can be calculated based on the difference between the total solid volume and the total system volume obtained from thermodynamic phase equilibrium calculations, such as... Figure 2 As shown, the degree of hydration and the amount of solid products generated can also be quantitatively characterized using relevant pore structure models. Since pore structure directly affects the distribution of internal defects, effective load-bearing area, and stress transfer capacity of materials, porosity is a crucial microstructural parameter influencing the compressive strength of alkali slag materials. When the system porosity is high, the continuous cementitious skeleton structure inside the material is disrupted, easily forming stress concentration areas, thus leading to a decrease in compressive strength.

[0044] Compared to using only macroscopic mix proportion parameters, porosity parameters can further characterize the degree of densification of the internal microstructure and the filling effect of hydration products. Therefore, introducing it into the machine learning model can further enhance the model's ability to characterize the evolution process of the internal structure of the material, improve the accuracy of compressive strength prediction, and enhance the model's ability to explain the mechanism.

[0045] In step (3), the basic mix proportion parameters, C-(N)-ASH gel content and porosity are input into the RF algorithm model.

[0046] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 Increased to 0.88.

[0047] Example 3 This embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Embodiment 2, except that: In step (2), the content of the hydrotalcite-like phase is further extracted as a thermodynamic characteristic parameter. The hydrotalcite-like phase is a common layered double hydroxide hydration product in alkaline slag material systems. It is typically generated by the reaction of elements such as Mg and Al in the slag under alkaline conditions, such as... Figure 2 , Figure 3 The content of hydroxyl-type hydrotalcite-OH in the brownish-red region. This phase can fix interlayer ions within the system and participate in the regulation of pore structure and stabilization of gel skeleton, thus having an important influence on the later densification of material structure and the development of mechanical properties.

[0048] Furthermore, the amount of hydrotalcite-like phase formed is closely related to the MgO content, alkali activation conditions, and curing environment in the system. Its changes can reflect, to some extent, the degree of hydration reaction and the evolution characteristics of phase composition within the alkali slag system. Therefore, inputting the hydrotalcite-like phase content as a thermodynamic characteristic parameter into a machine learning model can further enhance the model's ability to characterize the evolution of the material's internal microstructure, and improve the model's prediction accuracy and mechanistic explanation capabilities.

[0049] In step (3), the basic mix proportion parameters, C-(N)-ASH gel content, porosity and hydrotalcite phase content are input into the RF algorithm model.

[0050] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 Increased to 0.91.

[0051] Example 4 This embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Embodiment 3, except that: In step (2), OH is further extracted. -The concentrations of certain alkali metal ions, the proportion of unreacted slag, and the contents of AFt and AFm phases were used as thermodynamic characteristic parameters. Both AFt and AFm phases are important aluminate hydration products in alkali slag material systems. The AFt phase typically forms rapidly in the early stages of the system; its needle-like crystal structure can fill pores to some extent and promote early structure formation. However, as sulfate ions are gradually consumed, some AFt phase will further transform into AFm phase in the later stages. This transformation process is accompanied by changes in solid volume and pore structure reconstruction, which can easily lead to self-shrinkage and microcracks within the system, thus affecting the later strength development and structural stability of the material.

[0052] Therefore, the formation and transformation behavior of AFt and AFm phases can reflect the hydration reaction pathway and volume stability variation law within the alkali slag material system to a certain extent, and their content changes are strongly correlated with the evolution of the material's compressive strength. Inputting the contents of AFt and AFm-like phases as thermodynamic characteristic parameters into the machine learning model can further enhance the model's ability to characterize the later-stage structural stability and strength variation trends of the system, improving the model's prediction accuracy and mechanistic explanation capabilities.

[0053] Furthermore, the proportion of unreacted slag is also input into the machine learning model as a thermodynamic characteristic parameter. Because the actual degree of reaction of slag particles varies significantly under different alkali activation conditions, curing temperatures, and ages, the actual content of cementitious products generated within the system under the same slag dosage is not entirely the same. The reacted slag can further hydrate to generate hydration products with cementing effects, such as C-(N)-ASH gel, while unreacted slag exists more primarily as an inert particulate filler phase, and its contribution to the system strength differs significantly from that of the cementitious products.

[0054] Therefore, compared with the traditional method of using only macroscopic proportion parameters such as slag content for prediction, the introduction of the proportion of unreacted slag can further distinguish the difference between the "effective reaction part" and the "inert filling part" in the system, thereby more realistically reflecting the source of internal strength formation and the degree of hydration reaction, and improving the model's prediction accuracy and mechanism explanation ability.

[0055] OH - The concentration of OH can characterize the strength of the alkaline environment of the system to a certain extent, directly affecting the dissolution rate of the glassy phase of the slag and the release behavior of active components such as Si and Al, thereby further affecting the subsequent hydration product formation process and the development of system strength. Therefore, OH... - Introducing concentration as a thermodynamic characteristic parameter into the model helps to enhance the model's ability to characterize the system's reactivity and changes in the hydration environment.

[0056] In step (3), all thermodynamic characteristic parameters (C-(N)-ASH gel content, porosity, hydrotalcite phase content, OH-) are included. - The concentrations of alkali metal ions, the proportion of unreacted slag, and the content of AFt and AFm phases are input into the RF algorithm model along with the basic mix proportion parameters (slag chemical composition, activator type, activator modulus, alkali equivalent, and water-cement ratio).

[0057] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 Increased to 0.94.

[0058] Example 5 This embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Embodiment 4, except that: In step (4), the support vector machine (SVM) algorithm is used instead of the RF algorithm to establish a compressive strength prediction model.

[0059] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 It is 0.86.

[0060] Example 6 This embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Embodiment 4, except that: In step (4), the artificial neural network (ANN) algorithm is used instead of the RF algorithm to establish a compressive strength prediction model.

[0061] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 It is 0.90.

[0062] Example 7 This embodiment provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Embodiment 4, except that: In step (4), the XGBoost algorithm is used instead of the RF algorithm to establish a compressive strength prediction model. The learning rate is set to 0.05, the maximum tree depth is set to 6, and the subsampling rate is set to 0.8.

[0063] Referring to Table 1 below, the test results show that the prediction coefficient of determination R for the 28-day compressive strength of the model in this embodiment is... 2 It reached 0.95.

[0064] Comparative Example 1 This comparative example provides a method for predicting the compressive strength of alkali slag materials. The specific steps are basically the same as those in Example 7, except that: In steps (2) and (3), thermodynamic characteristic parameters are not used; only basic mix proportion parameters (slag chemical composition, activator type, activator modulus, alkali equivalent, and water-cement ratio) are used to train the XGBoost algorithm model.

[0065] Referring to Table 1 below, the test results show that the coefficient of determination R for predicting the 28-day compressive strength of this comparative model is... 2 It dropped to 0.79.

[0066] Comparative Example 2 This comparative example provides a method for predicting the compressive strength of alkali slag materials, the specific steps of which are as follows: The 28-day compressive strength of alkali slag material system was predicted by using the traditional multiple linear regression empirical formula and inputting basic mix proportion parameters (slag chemical composition, activator type, activator modulus, alkali equivalent, and water-cement ratio).

[0067] Referring to Table 1 below, the test results show that the coefficient of determination R for predicting the 28-day compressive strength of this comparative model is... 2 It is only 0.71.

[0068] The model prediction results and experimental test results of the above embodiments and comparative examples are summarized in Table 1 below.

[0069] Table 1. Summary of Model Prediction Results and Experimental Test Results for Each Embodiment and Comparative Example

[0070] As shown in the table above, with the increase of the number of thermodynamic characteristic parameters, the error between the model prediction results and the experimental test results gradually decreases, and the model determination coefficient R... 2 The gradual increase indicates that thermodynamic parameters can effectively improve the prediction accuracy of compressive strength in alkali slag material systems. Furthermore, under the condition of using all thermodynamic parameters, the XGBoost model has higher prediction accuracy compared to the support vector machine model and the artificial neural network model, with its coefficient of determination R0... 2 The accuracy of 0.95 indicates that the XGBoost model is more suitable for predicting the compressive strength of alkali slag material systems. Furthermore, the results of Comparative Examples 1 and 2 show that when thermodynamic parameters are not introduced or traditional empirical formulas are used, the model prediction error increases significantly, further demonstrating the effectiveness and superiority of the method of this invention.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation, characterized in that, Includes the following steps: Collect raw data on the alkali slag material system; The hydration reaction process of alkali slag material system under different curing ages was simulated and calculated based on thermodynamic simulation method to obtain thermodynamic characteristic parameters that can characterize the internal reaction state of the material. The calculation method for the thermodynamic characteristic parameters is as follows: the Cemdata18 thermodynamic database, customized for cementitious material systems, is used; based on the principle of minimum Gibbs free energy, an equilibrium criterion for minimizing Gibbs free energy in the alkali slag material system is established; the MPK kinetic model is used, combining three rate control mechanisms of nucleation and growth, diffusion, and shell chemical reaction, and the minimum value among the three rates is taken as the total reaction rate of the alkali slag material system. The degree of hydration of the slag at different curing ages is obtained by integrating the total reaction rate over time. Set the thermodynamic simulation temperature, use the actual amount of slag involved in the reaction at each age as the boundary condition, input it into the thermodynamic simulation software for quantitative phase equilibrium calculation, output the stacking evolution diagram of hydration products at different ages, and extract thermodynamic characteristic parameters from the evolution diagram. The basic mix proportion parameters and thermodynamic characteristic parameters in the original data are fused together, and after data preprocessing, a machine learning training dataset is constructed. A machine learning algorithm was used to train the training dataset to establish a predictive model for the compressive strength of alkali slag materials. Input the basic mix proportion parameters and corresponding thermodynamic characteristic parameters of the alkali slag material system to be predicted into the trained prediction model, and output the predicted compressive strength at the corresponding age.

2. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 1, characterized in that, The formula for the balance criterion is as follows: in, The total Gibbs free energy of the alkali slag material system; The total number of chemical components in the alkali slag material system; For the first The number of moles of each component; For the first The chemical potential of a component in its current state.

3. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 1, characterized in that, The formula for calculating the total reaction rate is as follows: in, The total reaction rate is... The degree of slag hydration, The reaction rate during the nucleation and growth control phases. To control the reaction rate during the diffusion phase, To control the reaction rate in the shell chemical reaction stage, For time.

4. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 3, characterized in that, The formula for calculating the reaction rate of the nucleation and growth control stages is as follows: The formula for calculating the reaction rate in the diffusion-controlled stage is as follows: in, This represents the reaction rate constant during the nucleation and growth control phases. The reaction rate constant during the diffusion-controlled phase, This is the shape index.

5. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 4, characterized in that, The formula for calculating the reaction rate in the shell chemical reaction control stage is as follows: in, The rate constant for the shell chemical reaction controlling the reaction phase. This is the mechanism index.

6. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 1, characterized in that, The thermodynamic simulation method employs one of the following software suites: GEMS, PHREEQC, or FactSag.

7. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 1, characterized in that, The thermodynamic characteristic parameters include: C-(N)-ASH gel content, porosity, hydrotalcite phase content, and OH content. - Concentration, concentration of some alkali metal ions, proportion of unreacted slag, and content of AFt and AFm phases.

8. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 1, characterized in that, The basic mix proportion parameters include: slag chemical composition, activator type, activator modulus, alkali equivalent, and water-cement ratio.

9. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 1, characterized in that, The machine learning algorithm used is one of the following: random forest algorithm, support vector machine algorithm, artificial neural network algorithm, and XGBoost algorithm.

10. The method for predicting the compressive strength of alkali slag materials based on thermodynamic simulation according to claim 9, characterized in that, The learning rate of the XGBoost algorithm is set to 0.05, the maximum tree depth is set to 6, and the subsampling rate is set to 0.8.

Images