Method for predicting gelation activity of alkali-activated tailings based on machine learning and strength of concrete prepared from alkali-activated tailings

Abstract

The invention discloses a method for predicting the gelation activity of alkali-activated tailings and the strength of concrete prepared from the alkali-activated tailings based on machine learning, and the method comprises the following steps: collecting heat flow data of typical alkali-activated tailings, and converting the heat flow data by using Fourier transform and discrete cosine transform functions; and establishing a prediction model of the gelling activity of the alkali-activated tailings and the strength of the tailings-based concrete based on a machine learning model. Compared with a traditional concrete strength prediction model, the method has the advantages that the hydration dynamic data of the alkali-activated tailings are converted, nonlinear features can be remarkably reduced, a data structure is simplified, and a machine learning model can traverse all data, so that the prediction model has higher prediction precision and reliability.

Description

Technical Field

[0001] The present invention belongs to the field of tailings solid waste resource utilization, and in particular relates to a method for predicting the alkali-activated tailings gelling activity and the strength of concrete prepared therefrom based on machine learning. Background Art

[0002] Tailings are waste generated during the mining process of mineral resources. They mainly come from the mineral particles of no economic value or low grade left after the ore is beneficiated. They are the main component of industrial solid waste. These wastes usually contain harmful substances. If they are not handled properly, they will pollute the soil and water sources and may cause geological disasters such as tailings dam collapse, posing a serious threat to the environment and human health. However, tailings often contain a certain amount of valuable elements and components that can be used in the production of building materials. Through appropriate reuse technology, it can not only reduce environmental pollution, but also achieve effective recovery and recycling of resources and promote sustainable development.

[0003] The main components of most tailings are oxides such as SiO2, Al2O3, and CaO, which are the same as the main components of cement or traditional cementitious materials. Therefore, tailings have the potential to replace cement cementitious materials. By stimulating their cementitious activity, the amount of cement clinker in concrete can be reduced, effectively reducing carbon emissions. Studies have shown that alkali activators such as sodium hydroxide, sodium silicate, sodium carbonate, and sodium sulfate can effectively improve the cementitious activity of tailings, but when used as adhesives, the solidification of concrete is slow and the compressive strength is low. These problems can be alleviated by forming additional hydration products under the composite addition of calcium-containing additives such as slag. Chinese patent ZL202211359391.7 discloses a calcium-rich alkali-activated material with adjustable setting time and a preparation method thereof. By activating tailings-high calcium aluminum silicate cementitious materials with sodium silicate, not only can the setting time of concrete be significantly extended, but also the strength can be improved.

[0004] Emarah et al. (Compressive strength analysis of fly ash-based geopolymer concrete using machine learning approaches, Results in Materials, 16 (2022) 100347) disclosed a machine learning method based on artificial neural networks (ANNs), deep neural networks (DNNs) and deep residual networks (ResNet) to predict the compressive strength of concrete. In the modeling process, 12 valid parameters were used as input variables. Due to the large number of input parameters and their mutual influence, the accuracy of the prediction model was not high. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problem that the activity of existing alkali-activated tailings cementitious materials is difficult to predict, and to provide a method for predicting the alkali-activated tailings cementitious activity based on machine learning. The method reduces the number of data sets by changing the data, uses heat flow data to characterize its cementitious activity, and uses a support vector machine (SVM) model to perform regression analysis on the heat flow data and concrete strength, thereby providing a method for predicting the strength of tailings-based concrete with high accuracy.

[0006] The present invention discloses a method for predicting the strength of concrete prepared by alkali-activated tailings gelling activity and the same based on machine learning, which is characterized by comprising the following steps: S1. Using an isothermal calorimeter to collect heat flow data of 50 different alkali-activated tailings-slag composite cementitious materials; S2, performing Fourier transform (FFT) on the collected heat flow data to obtain the converted corresponding data, including the real part and the imaginary part; S3, performing discrete cosine transform (DCT) function transformation on the Fourier transformed data; S4. Based on the machine learning model, the data after FFT and DCT conversion were trained by machine learning, and the 10-fold cross validation method and grid search method were used to determine the optimal hyperparameters, and the prediction model of basicity (n), silicon-sodium ratio (Ms), slag content (S) and heat flow data was established; S5, adding sand and gravel coarse aggregate and water to the above-mentioned cementitious material to prepare slurry, pouring and curing to obtain concrete, and testing the strength of the concrete; S6. Based on the machine learning model, machine learning training is performed on the concrete strength and cementitious material heat flow data to establish a regression prediction model between the cementitious material heat flow and concrete strength.

[0007] The tailings in the tailings-slag-alkali activator composite cementitious material are one or more of iron tailings, copper tailings, zinc tailings, manganese tailings and the like, the SiO2 content in the tailings is greater than 55wt%, the Al2O3 content is greater than 15wt%, and the particle size of the tailings is 20-50μm.

[0008] The slag in the tailings-slag-alkali activator composite cementitious material is blast furnace slag powder, and the particle size of the slag powder is 10-15 μm.

[0009] The tailings-slag-alkali activator composite gelling material contains 70-100 wt % of tailings and 0-30 wt % of slag.

[0010] The alkaline activator is one or more of sodium hydroxide, sodium silicate, sodium carbonate and sodium sulfate.

[0011] The amount of the alkali activator added is determined according to the alkalinity (n) and the silicon-sodium ratio (M s), n is the mass ratio of Na2O to the total powder, M s is the mass ratio of SiO2 to Na2O, where 0.025≤n≤0.1, 0.9≤Ms≤1.2.

[0012] The DCT function relationship is:

[0013] Where N is the number of data points, is the original data, is the Kronecker function, is the order, when =1, =1 if the value is set to 0, otherwise it is 0.

[0014] The 10-fold cross-validation method divides the dataset into ten subsets, trains the model on nine subsets, and tests it on the remaining subset, repeating the training ten times for each combination. The grid search method systematically traverses multiple potential optimal hyperparameter combinations, performing cross-validation in the process to determine the most accurate prediction model.

[0015] The machine learning model is one of a genetic algorithm neural network (GANN), a support vector machine (SVM), a random forest (RF) and a long short-term memory network (LSTM).

[0016] The mass ratio of the sand and gravel coarse aggregate, water and cementitious material is 6:1:2.

[0017] The present invention has the beneficial effects: (1) The hydration kinetics of alkali-activated tailings cementitious materials are relatively complex. The SiO2 and Al2O3 content in the tailings, the amount of slag added, the amount of alkali activator used, the water-cement ratio and other factors have a great influence on the activity of the tailings cementitious materials. It is difficult to produce reliable predictions directly using any machine learning model. In the present invention, the heat flow data during the hydration process of alkali-activated tailings cementitious materials is used to characterize its cementitious activity. The alkalinity, silicon-sodium ratio and slag addition data are all converted into heat flow data, realizing a direct mapping relationship between heat flow data and concrete strength. Therefore, the reliability of the prediction model is higher.

[0018] (2) In the present invention, FFT and DCT are used to transform the hydration dynamics data of cementitious materials, which can significantly reduce nonlinear characteristics and simplify the data structure, thereby ensuring that the machine learning model can traverse all data, which can not only reduce the degrees of freedom and complexity of the training set, but also improve the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1The predicted value and the true value of the concrete strength involved in the first embodiment of the present invention; Figure 2 The predicted value and the true value of the concrete strength involved in the second embodiment of the present invention; Figure 3 These are the predicted value and true value results of concrete strength involved in the third embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described below in conjunction with the examples. The following examples are only used to more clearly illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Embodiment 1

[0021] The method for predicting the gelling activity of alkali-activated tailings and the strength of concrete prepared therefrom based on machine learning is characterized by comprising the following steps: S1. Using an isothermal calorimeter to collect heat flow data of 50 different alkali-activated tailings-slag composite cementitious materials; S2, performing Fourier transform (FFT) on the collected heat flow data to obtain the converted corresponding data, including the real part and the imaginary part; S3, performing discrete cosine transform (DCT) function transformation on the Fourier transformed data; S4. Based on the random forest (RF) model, machine learning training was performed on the data after FFT and DCT conversion. The 10-fold cross-validation method and grid search method were used to determine the optimal hyperparameters, and a prediction model for alkalinity (n), silicon-sodium ratio (Ms), slag content (S) and heat flow data was established.

[0022] S5, adding sand and gravel coarse aggregate and water to the above-mentioned cementitious material to prepare slurry, pouring and curing to obtain concrete, and testing the strength of the concrete; S6. Based on the random forest (RF) model, machine learning training is performed on the concrete strength and cementitious material heat flow data to establish a regression prediction model between the cementitious material heat flow and concrete strength.

[0023] The tailings in the tailings-slag-alkali activator composite cementitious material are one or more of iron tailings, copper tailings, zinc tailings, manganese tailings and the like, the SiO2 content in the tailings is greater than 55wt%, the Al2O3 content is greater than 15wt%, and the particle size of the tailings is 20-50μm.

[0024] The slag in the tailings-slag-alkali activator composite cementitious material is blast furnace slag powder, and the particle size of the slag powder is 10-15 μm.

[0025] The tailings-slag-alkali activator composite gelling material contains 70-100 wt % of tailings and 0-30 wt % of slag.

[0026] The alkaline activator is one or more of sodium hydroxide, sodium silicate, sodium carbonate and sodium sulfate.

[0027] The amount of the alkali activator added is determined according to the alkalinity (n) and the silicon-sodium ratio (M s ), n is the mass ratio of Na2O to the total powder, M s is the mass ratio of SiO2 to Na2O, where 0.025≤n≤0.1, 0.9≤Ms≤1.2.

[0028] The DCT function relationship is:

[0029] Where N is the number of data points, is the original data, is the Kronecker function, is the order, when =1, =1 if the value is set to 0, otherwise it is 0.

[0030] The 10-fold cross-validation method divides the dataset into ten subsets, trains the model on nine subsets, and tests it on the remaining subset, repeating the training ten times for each combination. The grid search method systematically traverses multiple potential optimal hyperparameter combinations, performing cross-validation in the process to determine the most accurate prediction model.

[0031] The mass ratio of the sand and gravel coarse aggregate, water and cementitious material is 6:1:2.

[0032] Finally, we get the intensity regression R 2 The predicted value of concrete strength is compared with the actual value. Figure 1 shown. Embodiment 2

[0033] This embodiment is basically the same as the first embodiment, except that: (1) In step S4, machine learning training was performed on the data after FFT and DCT conversion based on the genetic algorithm neural network (GANN). The 10-fold cross-validation method and grid search method were used to determine the optimal hyperparameters, and a prediction model for alkalinity (n), silicon-sodium ratio (Ms), slag content (S) and heat flow data was established.

[0034] (2) In step S6, machine learning training is performed on the concrete strength and cementitious material heat flow data based on the genetic algorithm neural network (GANN) to establish a regression prediction model between the cementitious material heat flow and concrete strength.

[0035] Finally, we get the intensity regression R 2 The predicted value of concrete strength is compared with the actual value. Figure 2 shown. Embodiment 3

[0036] This embodiment is basically the same as the first embodiment, except that: (1) In step S4, machine learning training is performed on the data after FFT and DCT conversion based on the support vector machine (SVM) model. The 10-fold cross-validation method and grid search method are used to determine the optimal hyperparameters, and a prediction model for alkalinity (n), silicon-sodium ratio (Ms), slag content (S) and heat flow data is established.

[0037] (2) In step S6, machine learning training is performed on the concrete strength and cementitious material heat flow data based on the support vector machine (SVM) model to establish a regression prediction model between cementitious material heat flow and concrete strength.

[0038] Finally, we get the intensity regression R 2 The predicted value of concrete strength is compared with the actual value. Figure 3 shown.

[0039] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the gelling activity of alkali-activated tailings and the strength of concrete prepared therefrom based on machine learning, characterized in that: The steps include: S1. Using an isothermal calorimeter to collect heat flow data of 50 different alkali-activated tailings-slag composite cementitious materials; S2, performing Fourier transform (FFT) on the collected heat flow data to obtain the converted corresponding data, including the real part and the imaginary part; S3, performing discrete cosine transform (DCT) function transformation on the Fourier transformed data; S4. Based on the machine learning model, the data after FFT and DCT conversion were trained by machine learning, and the 10-fold cross validation method and grid search method were used to determine the optimal hyperparameters, and the prediction model of basicity (n), silicon-sodium ratio (Ms), slag content (S) and heat flow data was established; S5, adding sand and gravel coarse aggregate and water to the above-mentioned cementitious material to prepare slurry, pouring and curing to obtain concrete, and testing the strength of the concrete; S6. Based on the machine learning model, machine learning training is performed on the concrete strength and cementitious material heat flow data to establish a regression prediction model between the cementitious material heat flow and concrete strength.

2. The method for predicting the gelling activity of tailings using alkali-stimulated tailings based on machine learning and the strength of concrete prepared therefrom according to claim 1, characterized in that: The tailings in the tailings-slag-alkali activator composite cementitious material are one or more of iron tailings, copper tailings, zinc tailings, manganese tailings and the like, the SiO2 content in the tailings is greater than 55wt%, the Al2O3 content is greater than 15wt%, and the particle size of the tailings is 20-50μm.

3. The method for predicting the gelling activity of tailings using alkali-stimulated tailings and the strength of concrete prepared therefrom based on machine learning according to claim 1, characterized in that: The slag in the tailings-slag-alkali activator composite cementitious material is blast furnace slag powder, and the particle size of the slag powder is 10-15 μm.

4. The method for predicting the alkali-activated tailings gelling activity and the strength of concrete prepared therefrom based on machine learning according to claim 1, characterized in that: The tailings-slag-alkali activator composite gelling material has a tailings content of 70-100 wt %, and a slag content of 0-30 wt %.

5. The method for predicting the gelling activity of tailings and the strength of concrete prepared therefrom based on alkali-induced tailings based on machine learning according to claim 1, characterized in that: The alkaline activator is one or more of sodium hydroxide, sodium silicate, sodium carbonate and sodium sulfate.

6. The method for predicting the alkali-activated tailings gelling activity and the strength of concrete prepared therefrom based on machine learning according to claim 1, characterized in that: The amount of the alkali activator added is determined according to the alkalinity (n) and the silicon-sodium ratio (M s ), n is the mass ratio of Na2O to the total powder, M s is the mass ratio of SiO2 to Na2O, where 0.025≤n≤0.1, 0.9≤Ms≤1.

2.

7. The method for predicting the alkali-activated tailings gelling activity and the strength of concrete prepared therefrom based on machine learning according to claim 1, characterized in that: The DCT function relationship is: ； Where N is the number of data points, is the original data, is the Kronecker function, is the order, when =1, =1 if the value is set to 0, otherwise it is 0.

8. The method for predicting the gelling activity of tailings activated by alkali based on machine learning and the strength of concrete prepared therefrom according to claim 1, characterized in that: The machine learning model is one of a genetic algorithm neural network (GANN), a support vector machine (SVM), a random forest (RF) and a long short-term memory network (LSTM).

9. The method for predicting the gelling activity of tailings and the strength of concrete prepared therefrom based on alkali-induced tailings based on machine learning according to claim 1, characterized in that: The 10-fold cross-validation method divides the dataset into ten subsets, trains the model on nine subsets, and tests it on the remaining subset, repeating the training ten times for each combination. The grid search method systematically traverses multiple potential optimal hyperparameter combinations, performing cross-validation in the process to determine the most accurate prediction model.

10. The method for predicting the gelling activity of tailings and the strength of concrete prepared therefrom based on alkali-induced tailings based on machine learning according to claim 1, characterized in that: The mass ratio of the sand and gravel coarse aggregate, water and cementitious material is 6:1:2.

Images