Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Compressive strength prediction and low-carbon optimization of fly ash geopolymer concrete based on big data and ensemble learning

  • Peiling Jiang,

    Roles Funding acquisition, Investigation, Methodology, Validation, Visualization, Writing – original draft

    Affiliation Zhejiang Tongji Vocational College of Science and Technology, Zhejiang, China

    ⨯
  • Diansheng Zhao,

    Roles Funding acquisition, Investigation, Writing – original draft

    Affiliation Zhejiang University of Technology, Zhejiang, China

    ⨯
  • Cheng Jin,

    Roles Investigation

    Affiliation Zhejiang University of Technology Engineering Design Group Co. Ltd, Zhejiang, China

    ⨯
  • Shan Ye,

    Roles Investigation, Writing – original draft

    Affiliation Zhejiang Tongji Vocational College of Science and Technology, Zhejiang, China

    ⨯
  • Chenchen Luan ,

    Roles Conceptualization, Writing – review & editing

    luanchenchen@hit.edu.cn

    Affiliations Institute of Advanced Study, Chengdu University, Chengdu, China, School of Civil and Environmental Engineering, Harbin Institute of Technology, Shenzhen, China

    ⨯
  • Rana Faisal Tufail

    Roles Writing – review & editing

    Affiliation Civil Engineering Department, Wah Campus, COMSATS University Islamabad, Rawalpindi, Pakistan

    ⨯

Abstract

Portland cement concrete (PCC) is a major contributor to human-made CO2 emissions. To address this environmental impact, fly ash geopolymer concrete (FAGC) has emerged as a promising low-carbon alternative. This study establishes a robust compressive strength prediction model for FAGC and develops an optimal mixture design method to achieve target compressive strength with minimal CO2 emissions. To develop robust prediction models, comprehensive factors, including fly ash characteristics, mixture proportions, curing parameters, and specimen types, are considered, a large dataset comprising 1136 observations is created, and polynomial regression, genetic programming, and ensemble learning are employed. The ensemble learning model shows superior accuracy and generalization ability with an RMSE value of 1.81 MPa and an R2 value of 0.93 in the experimental validation set. Then, the study integrates the developed strength model with a life cycle assessment-based CO2 emissions model, formulating an optimal FAGC mixture design program. A case study validates the effectiveness of this program, demonstrating a 16.7% reduction in CO2 emissions for FAGC with a compressive strength of 50 MPa compared to traditional trial-and-error design. Moreover, compared to PCC, the developed FAGC achieves a substantial 60.3% reduction in CO2 emissions. This work provides engineers with tools for compressive strength prediction and low carbon optimization of FAGC, enabling rapid and highly accurate design of concrete with lower CO2 emissions and greater sustainability.

Citation: Jiang P, Zhao D, Jin C, Ye S, Luan C, Tufail RF (2024) Compressive strength prediction and low-carbon optimization of fly ash geopolymer concrete based on big data and ensemble learning. PLoS ONE 19(9): e0310422. https://doi.org/10.1371/journal.pone.0310422

Editor: Afzal Husain Khan, Jazan University, SAUDI ARABIA

Received: March 4, 2024; Accepted: August 31, 2024; Published: September 12, 2024

Copyright: © 2024 Jiang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the manuscript and its Supporting information files.

Funding: We received funding from Zhejiang Province Construction Scientific Research Project, grant number 2022K242. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Abbreviations: c, Specific Heat Capacity; C/FA, Coarse Aggregate to Fly Ash Mass Ratio; Di, ith Concrete Component’s Transport Distance; ECuring, CO2 Emissions from Concrete Curing; EngCuring, Energy Use of Concrete Curing; EngHeat, Energy Use for Heating; EngLoss, Energy Use for Offsetting Heat Loss; ERawMatProd, CO2 Emissions from Production of Concrete Raw Materials; ERawMatTrans, CO2 Emissions from Transport of Concrete Raw Materials; F/FA, Fine Aggregate to Fly Ash Mass Ratio; FAGC, Fly Ash Geopolymer Concrete; fc,pred(x), Compressive Strength Prediction; fc,target, Target Compressive Strength; FEng, CO2 Emission Factor of Energy; Fi, ith Concrete Component’s CO2 Emission Factor; FTrans, CO2 Emission Factor of Transport; H2O/FA, Water to Fly Ash Mass Ratio; LCA, Life Cycle Assessment; M, Concrete Mass; m, Number of Dataset Sample; m20, Number of Samples with Prediction-to-Actual Value Ratio between 0.80 and 1.20; MCoarseAggregate, Coarse Aggregate Mass in 1 m3 FAGC; MFineAggregate, Fine Aggregate Mass in 1 m3 FAGC; MFlyAsh, Fly Ash Mass in 1 m3 FAGC; MH2O,NS, Water Mass in NaOH Solution; MH2O,WG, Water Mass in Water Glass; MNaOH, NaOH Mass in 1 m3 FAGC; MNaOHSolution, NaOH Solution Mass in 1 m3 FAGC; MWaterglass, Water Glass Mass in 1 m3 FAGC; Na/Al, Sodium to Aluminum Molar Ratio; P, Heating Power for Offsetting Heat Loss; P, Heating Power for Offsetting Heat Loss at Temperature of 80°C; PAl2O3,FA, Mass percent of Al2O3 in Fly Ash; PCaO,FA, Mass percent of CaO in Fly Ash; PCC, Portland Cement Concrete; PFe2O3,FA, Mass percent of Fe2O3 in Fly Ash; PNa2O,WG, Mass percent of Na2O in Water Glass; PNaOH,NS, Mass percent of NaOH in NaOH Solution; PSiO2,FA, Mass percent of SiO2 in Fly Ash; PSiO2,WG, Mass percent of SiO2 in Water Glass; R2, Coefficient of Determination; RMSE, Root Mean Squared Error; Si/Al, Silicon to Aluminum Molar Ratio; T, Curing Temperature; T0, Ambient Temperature; t1, Heat-Curing Time; t2, Total Curing Time; Wi, ith Component’s Weight in 1 m3 Concrete; y, Actual Value (Observation) of Sample; ymean, Mean Value of Observations; ypred, Prediction of Sample; α, Air Content; ρC, Density of Coarse Aggregate; ρF, Density of Fine Aggregate; ρFA, Density of Fly Ash; ρNS, Density of NaOH Solution; ρWG, Density of Water Glass

1. Introduction

1.1. Background

Portland cement is the essential component of concrete, the most consumed material in the world after water; However, it accounts for 8% of human-made CO2 emissions, contributing significantly to climate change [1, 2]. To address this, geopolymer, also known as low-calcium alkali-activated material, has gained worldwide attention as a promising low-carbon substitute for cement [1, 2]. It is synthesized using industrial by-products or waste ashes containing glassy/amorphous Si and Al with a small alkali-activator. Fly ash, a prevalent by-product of coal combustion, is an ideal precursor for the geopolymer. It is an excellent source of amorphous SiO2 and Al2O3. Meanwhile, it is also the largest industrial by-product in the world. With global coal combustion generating one billion tons of fly ash annually, of which only 30% is utilized [3], the potential for fly ash geopolymer concrete (FAGC) to mitigate CO2 emissions is substantial. Compared with conventional Portland cement concrete (PCC), FAGC not only reduces CO2 emissions [4, 5] but also exhibits comparable engineering properties [6, 7], better durability in sulfates or acids [8], and higher fire resistance [9, 10].

A key issue for concrete, including but not limited to FAGC, is the mixture design optimization, i.e., determining the optimal mixture proportions to yield a concrete that meets performance targets, such as compressive strength, while minimizing CO2 emissions [11, 12]. In response to the challenges posed by climate change, the focus of concrete research today has shifted from enhancing concrete performance to reducing its environmental impact while ensuring performance [13]. While traditional trial batching (experimental design) is one approach to mixture design optimization, it involves conducting numerous trials with various mixture parameters as variables. As the number of variables increases, the required number of trials grows exponentially, making the process time-consuming and resource-intensive. Additionally, this approach only gives relatively well-performing, rather than truly optimal, concrete mixtures.

To overcome the limitations of experimental design, modeling the relationships between concrete properties (such as compressive strength) and mixture proportions enables computational design. This mathematical approach can save time, labor, and resources, and potentially identify the truly optimal mixture proportions [11, 12, 14]. However, due to the highly nonlinear and complex nature of the relationships between concrete properties and mixture proportions, they cannot be captured by simple equations. In such cases, machine learning techniques offer valuable solutions to model complex relationships, provided a representative database of experimental results is available. Detailed and in-depth state-of-the-art reports on the engineering application of machine learning techniques are well-documented in the literature [15–18]. The successful use of machine learning in modeling cement concrete properties has also been extensively covered [19–22]. Furthermore, FAGC, as a highly promising low-carbon concrete, has attracted increasing interest in developing compressive strength prediction models using machine learning techniques [23–25].

Despite the advantage of machine learning in modeling complex relationships, the risk of overfitting remains a significant challenge [21, 26]. Overfitting means that although a model performs exceptionally well on the training data, the model gives inaccurate predictions for new data. Preventing overfitting in machine learning is crucial, and several strategies exist to address this issue. One effective strategy is to increase the size of the training dataset. A dataset exceeding 1000 observations is recommended [18]. However, most prior studies [27–36] on modeling the compressive strength of FAGC are limited to small data sets with 100–500 observations, as shown in Fig 1. Another approach to reducing overfitting risk is the use of ensemble learning methods, such as stacking. Compared with the use of a single machine learning model, stacking ensemble learning combines the prediction results of multiple models through weighting or other methods to produce a final prediction, usually achieving higher accuracy and better generalization ability, i.e., better predictive ability for unknown data. Asteris et al. [26] were the first to apply stacking ensemble learning to predict the compressive strength of concrete materials. They used four conventional machine learning techniques and combined the outputs using an artificial neural network to create the stacking ensemble model, which outperformed the individual models. Later, Li and Song [37] applied a similar approach to predict the compressive strength of PCC with rice husk ash, developing decision tree and extreme gradient boosting models and then combining outputs using linear regression. Their stacking ensemble model achieved higher accuracy and better performance on new datasets compared to single models.

thumbnail
Download:
Fig 1. Comparison of data volume in different studies.

https://doi.org/10.1371/journal.pone.0310422.g001

In summary, despite the progress made in developing compressive strength prediction models for FAGC based on mixture proportions, there is currently a lack of studies that utilize large datasets and ensemble learning methods. Moreover, while compressive strength prediction models are essential, they are only part of the broader objective of mixture design optimization. A tool for mixture design optimization of FAGC is still lacking. Besides compressive strength prediction models, mixture design optimization also requires modeling the relationship between CO2 emissions and mixture proportions and solving the concrete mixture design optimization problem [11, 12].

Considering these points, this study leverages a dataset comprising 1136 observations and employs an ensemble machine learning method to develop a robust compressive strength prediction model for FAGC. Then, based on the strength model and life cycle assessment (LCA)-based CO2 emissions model, this study develops an optimal FAGC mixture design program to design the FAGC mixture that satisfies the target strength and minimizes CO2 emissions. Finally, a case study is presented to evaluate the effectiveness of the optimal mixture design tool.

1.2. Research significance and novelty

To address the global challenges of climate change, FAGC has been widely considered a promising low-carbon substitute for conventional concrete. This study addresses the critical need for computational mixture design optimization to develop FAGC that meets key performance targets, such as compressive strength, while minimizing CO2 emissions. Achieving this goal requires a compressive strength prediction model with good generalization ability. Despite many studies on developing related models, there is currently a lack of studies using large datasets and ensemble learning methods—two crucial strategies for enhancing generalization. This study addresses this gap by using a dataset comprising 1136 observations and employing an ensemble machine learning method to develop a compressive strength prediction model for FAGC. To the author’s best knowledge, the database reported in this paper is the largest to date. The model exhibits superior accuracy and generalization ability in testing and validation datasets. Beyond merely predicting compressive strength, this study integrates this strength model and an LCA-based CO2 emissions model and develops a mixture design optimization tool to design the FAGC mixture that satisfies the target strength and minimizes CO2 emissions. The results of this work provide engineers with tools for compressive strength prediction and low-carbon optimization of FAGC, enabling rapid and highly accurate design of concrete with lower CO2 emissions and greater sustainability.

2. Materials and methods

2.1. Research process

Fig 2 illustrates the research process flowchart in this study, encompassing variable identification, data collection and preprocessing, modeling, experimental validation, and optimal mixture design. The study begins by identifying the input variables for modeling (Section 2.2). Subsequently, only data that provides information on all identified input variables is collected from the literature, resulting in a big dataset with 1136 observations (Section 2.3). Data preprocessing is a critical step, as it directly affects the performance and accuracy of the models (Section 2.4). Polynomial regression and genetic programming are employed to develop the prediction model, with outputs combined using linear regression to form the stacking ensemble model (Section 2.5). The dataset is randomly split into training and testing sets, with 75% of the data used for training, and the remaining 25% reserved for testing the models’ performance. Additionally, new experimental data is used to further assess the proposed models (Section 2.6). A comparison of the proposed models is conducted using performance indices (Section 2.7), with the best-performing model selected. Finally, an optimal mixture design method is developed, aiming to achieve target compressive strength while minimizing CO2 emissions, based on the strength model and an LCA-based CO2 emissions model (Section 2.8). The objective function is to minimize CO2 emissions, with achieving the desired compressive strength as a primary constraint, alongside other constraints (Section 2.9). The optimal mixture design is determined by solving the optimization problem defined by the objective function and constraints, using the scipy.optimize.minimize algorithm within Python’s optimization toolbox.

thumbnail
Download:
Fig 2. Research process flowchart.

https://doi.org/10.1371/journal.pone.0310422.g002

2.2. Variable identifying

This section discusses the factors influencing the compressive strength of FAGC and identifies the input variables in modeling. Table 1 compares the influencing factors considered in previous studies. The influencing factors include fly ash characteristics, mixture proportions, curing parameters, and specimen type. Previous studies only considered part of the influencing factors. To develop a robust compressive strength prediction model of FAGC, this study considers the fly ash characteristics, mix proportions, curing parameters, and specimen type together. The detailed parameters considered in this study include SiO2, Al2O3, Fe2O3, and CaO percent in fly ash (PSiO2,FA, PAl2O3,FA, PFe2O3,FA, PCaO,FA), molar ratios of Na to Al (Na/Al) and Si to Al (Si/Al), mass ratios of H2O to fly ash (H2O/FA), coarse aggregate to fly ash (C/FA), and fine aggregate to fly ash (F/FA), heat-curing temperature (T), heat-curing time (t1), and total curing time (t2). These parameters cover fly ash characteristics, mixture proportions, and curing parameters. Besides, this study also considers the effect of specimen shape by converting cylinder strength to cube strength before modeling [38].

thumbnail
Download:
Table 1. Comparison of influencing factors considered in different studies.

https://doi.org/10.1371/journal.pone.0310422.t001

2.2.1. Characteristics of fly ash.

Fly ash is a by-product and its properties can vary significantly. This variation in the physicochemical properties of fly ash from different sources can lead to substantial differences in the compressive strength for the same mixture proportion [39]. Therefore, it is crucial to consider the characteristics of fly ash in the model. The oxide composition of fly ash, particularly the percentages of SiO2, Al2O3, CaO, and Fe2O3, plays a crucial role in determining the characteristics of fly ash. These oxides make up approximately 90% of the composition, while the remaining oxides are negligible. The content of CaO and the total content of SiO2, Al2O3, and Fe2O3 are used to classify fly ash in ASTM C618-19. These oxides have a significant impact on the compressive strength of FAGC. Si and Al play major roles in geopolymer formation. The general formula of geopolymer is Nam[–(SiO2)q–AlO2]m·wH2O (usually abbreviated as N-A-S-H), where q is Si/Al (1, 2, or 3) [40, 41]. SiO2 and Al2O3 are dissolved in an alkaline medium and then re-polymerize to form inorganic polymers with a three-dimensional network structure formed by connecting SiO2 tetrahedra and AlO2- tetrahedra. Recent studies reported that Fe2O3 participates in geopolymerization like Al2O3 [42, 43]. The presence of CaO leads to the formation of C-A-S-H, which accelerates the hardening of geopolymers and promotes early high strength at room temperature [44–46]. Based on the above, this study considers the SiO2, Al2O3, Fe2O3, and CaO percent in fly ash as input variables. Other fly ash characteristics, such as the amorphous content, fineness, and density, also influence the strength. However, as most papers only provide the chemical components of fly ash without further details, this model focuses on the oxide components of fly ash as input variables. It may affect the applicability of the model to a certain extent.

2.2.2. Mixture proportions.

FAGC is produced by mixing coarse and fine aggregates, fly ash, and alkali-activated solution. The alkali-activated solution comprises Na2O, SiO2, and H2O, and is usually prepared by mixing NaOH solution and Na2SiO3 solution (water glass). The water-reducing admixture is not involved because the geopolymer system lacks effective water-reducing admixture [47]. Therefore, the mixture proportions of FAGC involve mass ratios of Na2O to fly ash (Na2O/FA) and SiO2 to fly ash (SiO2/FA) as well as H2O/FA, C/FA, and F/FA. This study uses Na/Al and Si/Al instead of Na2O/FA and SiO2/FA because the variations of compressive strength with Na2O/FA and SiO2/FA are essentially induced by the variations of Na/Al and Si/Al. Na is from the alkali-activated solution. Al is from fly ash. Si is from both the alkali-activated solution and fly ash. The values of these variables are calculated using Eqs (1–8). (1) (2) (3) (4) (5) (6) (7) (8) where MK is the mass of K, and PN,K is the mass percent of N in K. For example, MWaterglass is the mass of water glass, and PNa2O,WG is the mass percent of Na2O in water glass. 60, 62, 80, and 102 are the relative molecular mass of SiO2, Na2O, NaOH, and Al2O3.

2.2.3. Curing parameters.

Curing parameters are widely considered key factors affecting the performances of FAGC [23, 24, 48]. Two-stage curing, i.e., heat curing (e.g., 80°C for 24 hours) followed by room temperature curing, is usually used for FAGC. The strength of FAGC develops slowly at room temperature (about 20°C). Achieving sufficient strength requires at least 56 days [49]. Therefore, slightly-elevated temperature curing for a short time is usually employed to accelerate strength development. The curing process can be described using three variables, namely heat-curing temperature (T), heat-curing time (t1), and total curing time (t2). Herein, t1 = t2 means that only heat curing is adopted, and t1 = 0 means that only room temperature curing is adopted.

2.2.4. Specimen shape.

Specimen shape needs to be considered, while it is ignored in the previous studies. Researchers from various regions use different specimen shapes for concrete compressive strength testing because the regional standards vary. In contrast to Chinese and European standards, which favor concrete cubes, concrete cylinders with a length-to-diameter ratio of 2 are used in the American standard. The compressive strength results of concrete cubes and cylinders produced by the same batch are different. Cubes have higher compressive strength than cylinders. In this study, the effect of specimen shape is taken into account by converting the cylinder strength to cube strength before modeling. To obtain cube compressive strength, the cylinder compressive strength needs to multiply the factor of 1.18 if it is less than 50 MPa; else, it needs to multiply the factor of 1.04 [38].

2.3. Data collection

Only data that provides information on all identified input variables is collected from the literature, resulting in a big dataset with 1136 observations for model development. Table 2 displays part of the dataset. There is too much data to display in full. Please refer to S1 Table for all the data. Table 3 shows the statistical parameters of the dataset. The dataset is randomly split into training and testing sets, with 75% (shown in S2 Table) used for training, and the remaining 25% (shown in S3 Table) for testing. The dataset is created as follows: The literature pertaining to the compressive strength of FAGC are retrieved on the Web of Science and Google Scholar. Afterward, only the literature that offers the key information, such as the SiO2, Al2O3, CaO, Fe2O3 percent in fly ash, the dosage of fly ash, NaOH solution, Na2SiO3 solution, extra added H2O, coarse aggregate and fine aggregate, the concentration or mass fraction of NaOH solution, the Na2O and SiO2 content in Na2SiO3 solution, heat-curing temperature (T), heat-curing time (t1), total curing time (t2), the specimen type for strength test, and the compressive strength (fc), are collected [28, 45, 50–129]. Finally, the key information is transcribed into a spreadsheet.

thumbnail
Download:
Table 2. Dataset.

https://doi.org/10.1371/journal.pone.0310422.t002

thumbnail
Download:
Table 3. Statistical parameters of the data set.

https://doi.org/10.1371/journal.pone.0310422.t003

2.4. Data pre-processing

This section presents data pre-processing steps, including normalizing variables and ensuring that the data is in the right distribution. Before modeling, variables should approximate a normal distribution, which can be achieved by taking logarithms. Normalization is also required to scale variables to the same order of magnitude. Table 4 shows the relationship between input variables and model parameters.

thumbnail
Download:
Table 4. Relationship between variables and model parameters.

https://doi.org/10.1371/journal.pone.0310422.t004

2.5. Modeling techniques

This section presents the basic principles of machine learning techniques used in this work, polynomial regression, genetic programming, and ensemble learning.

2.5.1. Polynomial regression.

Based on Taylor expansion, a quadratic function can effectively approximate any nonlinear function. Eq (9) presents an example of the quadratic function involving three input variables, namely x0, x1, and x2. Polynomial regression is essentially multiple linear regression.

(9)

There are 12 input variables in this study. The model initially considers 90 independent variables, namely x0, x02, x0x1, …, x0x11, x1, x12, x1x2, …, x11 and x112. However, the correlation between the dependent variable and each selected independent variable is tested using t-test. The independent variable with a P-value of less than 0.05 is considered to have a weak correlation with the dependent variable, and it is deleted. Then, the model is re-solved based on the remaining variables.

2.5.2. Genetic programming.

Genetic programming is proposed based on the Theory of Biological Evolution. A formula population is built by randomly combining mathematical operations (+,–, ÷, *(×)), constants, and variables. Then, the fittest individuals are selected based on the fitness (RMSE in this study), and the next generation evolves from them by genetic operations, including crossover, mutation and copy. The iteration continues until the termination condition is met (obtaining a satisfactory formula or exhausting the pre-set iteration number). This study uses an open-source genetic programming “gplearn” based on Python. More detailed information on "gplearn" can be found in its official documentation [130]. It is noted that the division in “gplearn” is specifically defined to avoid dividing by zero. If the dividend a is less than 0.001, the result of dividing b by a is 1. In other cases, the division in “gplearn” is the same as the one we normally use. The goal is to obtain the laws of these factors affecting concrete based on data. The open-source programs provide efficient and effective tools for non-machine learning professionals, such as concrete designers or researchers. They do not need to write machine learning programs themselves but can leverage existing tools to analyze and apply the discovered patterns and laws from machine learning.

2.5.3. Ensemble learning.

Firstly, multiple machine learning models are trained. Then, the outputs of the previously trained models are used as inputs to train the ensemble model. In this step, linear regression is commonly used as the training method. For example, Polynomial regression model: fc = f(x); Genetic programming model: fc = g(x); ensemble model: fc = Af(x)+Bg(x). The coefficients A and B can be determined by linear regression.

2.6. Experimental verification

This study uses new experimental data for validating the prediction model of the compressive strength of FAGC. Fly ash from Shenzhen Daote Technology Co., Ltd is used in the experiment, and its oxide composition determined by X-ray fluorescence (PANalytical Axios, the Netherlands) is shown in Table 5. The alkali-activated solution is mixed of sodium silicate solution (Na2O = 8.83%; SiO2 = 27.64%; H2O = 63.53%) with sodium hydroxide (NaOH) solid (Purity > 98%) and tap water, and it is prepared 24 hours in advance before use. River sand with a fineness modulus of 2.5 is used as fine aggregate. Granite crushed stone is used as coarse aggregate. The particle sizes for the coarse aggregate are distributed as follows: 5–10 mm (35%), 10–16 mm (35%), and 16–20 mm (30%), respectively.

thumbnail
Download:
Table 5. Oxide composition of fly ash (%).

https://doi.org/10.1371/journal.pone.0310422.t005

The mixture proportion is shown in Table 6. A basic scenario (PSiO2,FA = 0.5741, PAl2O3,FA = 0.2282, PCaO,FA = 0.0778, PFe2O3,FA = 0.0622, Na/Al = 0.5, Si/Al = 2.5, H2O/FA = 0.35, C/FA = 3, F/FA = 1.5, T = 80°C, t1 = 1 day, t2 = 28 days) is set, and Na/Al, Si/Al, H2O/FA, C/FA, F/FA, T, t1, and t2 are individually changed to show their effects on the compressive strength of FAGC. The sample preparation and strength testing follow the Chinese standard GB/T 50081–2019 [131], which is equivalent to the European standard EN 12390–3 [132]. The compressive strength is tested on a 2000 KN electro-hydraulic mechanical testing machine (Changchun testing machine factory, Changchun, China), and cubes with a size of 100 mm are used for strength testing.

thumbnail
Download:
Table 6. Mixture proportion for experimental verification.

https://doi.org/10.1371/journal.pone.0310422.t006

2.7. Performance indices

The indices for evaluating the model are a root mean squared error (RMSE) computed using Eq (10) and a coefficient of determination (R2) computed using Eq (11), which measure the difference between the prediction and experiment. A low value of RMSE and a high value of R2 indicate good model performance. (10) (11) where y is the actual value (observation), ypred is the prediction, m is the number of the dataset sample, and ymean is the mean value of the observations.

In addition to RMSE and R2, the a20-index, computed using Eq (12), is a recently proposed performance index and has been used in many studies [26, 133–135]. (12) where m20 is the number of samples with a prediction-to-actual value ratio between 0.80 and 1.20. A high a20-index value indicates good model performance, with a value of 1 representing a perfect predictive model. The proposed a20-index has practical engineering significance, as it quantifies the proportion of samples where predictions fall within a ±20% deviation from the actual values.

2.8. Model for CO2 emissions

Eq (13) gives a CO2 emissions model for 1 m3 FAGC [136–139] based on the life cycle assessment (LCA) from gate to cradle. The system boundary is shown in Fig 3. The CO2 emissions of concrete mixing and casting are assumed to be negligible.

thumbnail
Download:
Fig 3. System boundary of CO2 emissions assessment.

https://doi.org/10.1371/journal.pone.0310422.g003

(13) where ERawMatProd is the CO2 emissions in the production of concrete raw materials (kgCO2/m3), ERawMatTrans is the CO2 emissions in the transport of concrete raw materials (kgCO2/m3), ECuring are the CO2 emissions in the concrete curing (kgCO2/m3), Fi is the ith concrete component’s CO2 emission factor (kgCO2/kg), Wi is the ith component’s weight in 1 m3 concrete (kg/m3), FTrans is the CO2 emission factor of transport (kgCO2/kg-km), Di is the ith component’s transport distance (km), FEng is the CO2 emission factor of the energy used, such as electricity, natural gas, solar energy, and so on (kgCO2/kWh), and EngCuring is the energy use of concrete curing (kWh).

Eq (14) is used for estimating the energy use of concrete curing [137, 139]. This equation is derived from the energy conservation. All the energy is utilized for heating the concrete to the curing temperature and offsetting heat loss. (14) where EngHeat is the energy used for heating the concrete to the curing temperature, EngLoss is the energy used for offsetting heat loss, M is the concrete mass (kg), c is the specific heat capacity of FAGC, assumed to be 700 J/kg°C [137], T is the heating-curing temperature (°C), T0 is the ambient temperature (°C), P is the heating power for offsetting heat loss (kW), and t1 is the heat-curing time (day).

According to the heat transfer theory, the heating power for offsetting heat loss P is directly proportional to the temperature difference between T and T0 [139]. Obviously, P at T0 is 0. Defining P at T of 80°C as P0, Eq (15) is derived to calculate P at different curing temperatures.

(15)

In summary, the CO2 emissions model, Eq (13), is actually a function of the mixture and curing parameters x, CO2 emissions = f(x).

2.9. Constraints for optimal mixture design

The optimal mixture design is determined by solving the optimization problem defined by the objective function and constraints. The objective function is to minimize CO2 emissions. The primary constraint is achieving the desired compressive strength, along with other constraints, as follows:

  1. 1) Compressive strength constraint: the compressive strength must not be less than the target compressive strength fc,target, (16)
    where fc,pred(x) is the compressive strength prediction, which is actually a function of the mixture and curing parameters x.
  2. 2) Volume constraint: the volume fraction of each raw material and the air content adds up to one,(17)
    where ρi is the density of the ith raw material (kg/m3). α is the air content, which is usually assumed to 0.01. The density of NaOH solution ρNS varies with the mass fraction of NaOH solution PNaOH,NS. The relationship between ρNS and PNaOH,NS is shown in Fig 4.
  3. 3) Workability constraint: H2O/FA must not be less than the required minimum value to ensure workability. Similarly, many studies reported that the fluidity sharply decreases when the aggregate volume increases to over 70% [117, 140]. Therefore, aggregate volume must not exceed 70% to ensure workability.
  4. 4) Non-negativity constraint: The mixture and curing parameters, as well as the mass fraction of NaOH solution, must not be less than 0.
  5. 5) Temperature and time constraints: The heat-curing temperature must be between 40°C and 120°C. And the heat-curing time must be between 1/6 and 4 days. These bounds are based on the observed range of values in the training set.
thumbnail
Download:
Fig 4. Relationship between density ρNS and mass fraction PNaOH,NS of NaOH solution.

https://doi.org/10.1371/journal.pone.0310422.g004

3. Results and discussion

3.1. Prediction models for compressive strength

S1 Appendix shows the prediction models of the compressive strength of FAGC by polynomial regression, genetic programming, and ensemble learning, respectively. Fig 5 compares the testing set’s observations and the three models’ predictions. The ensemble learning model shows higher accuracy than polynomial regression model and genetic programming model in the testing set. The RMSE value decreases from 7.15 MPa to 7.03 MPa, the R2 value increases from 0.73 to 0.74, and the a20-index value increases from 0.687 to 0.704 after using ensemble learning model.

thumbnail
Download:
Fig 5. Comparison between the testing set’s observations and the three model’s predictions.

https://doi.org/10.1371/journal.pone.0310422.g005

This study uses new experimental data to validate the prediction model of the compressive strength of FAGC. The experimental data (validation set) is given in S4 Table. Fig 6. compares the experimental results and the predicted results of the three models. The lowest RMSE value (1.81 MPa), the highest R2 value (0.93) and the highest a20-index value (1) indicate that the ensemble learning model successfully predicts the compressive strength of FAGC using a new fly ash. The ensemble learning model not only has a higher accuracy than the single model, but also shows better generalization ability to the new dataset.

thumbnail
Download:
Fig 6. Comparison between the validation set’s observations and the three model’s predictions.

https://doi.org/10.1371/journal.pone.0310422.g006

Preventing overfitting in machine learning is essential for developing models that generalize well to new data. In this study, overfitting is expected to be mitigated by collecting more data and using an ensemble of multiple model predictions. Performance evaluations based on testing and experimental validation datasets demonstrate the success of these strategies in preventing overfitting. The resultant model demonstrates strong generalization to new data.

3.2. Effects of mix proportions and curing parameters on compressive strength

The effects of mix proportions and curing parameters on compressive strength of FAGC are shown in Fig 7, including both experimental data and model estimation. Although there are some deviations between the experimental data and model estimation, the overall trends remain consistent as follows:

  1. 1) Compressive strength initially increases with an increase in Na/Al, but then decreases with a further increase in Na/Al. In N-A-S-H, the positive charge of Na+ balances the negative charge resulting from the substitution of Al into a Q4 molecular coordination, and NaOH will be excess if Na/Al is greater than one where Al only refers to amorphous Al and does not include crystalline Al [141]. The excess NaOH is detrimental to N-A-S-H formation, and the strength decreases with the increase of Na/Al when Na/Al is greater than 1. The Al here is the whole Al, including amorphous Al and crystalline Al, and the actual optimal Na/Al value is less than 1.
  2. 2) Similarly, compressive strength initially increases with an increase in Si/Al, but decreases with a further increase. The saturation of Si in the reaction system due to increased Si from Na2SiO3 solution retards the dissolution of fly ash, resulting in minimal increase or even decrease in compressive strength. The optimal Si/Al is about 2.5 [142–144].
  3. 3) Compressive strength decreases linearly with an increase in H2O/FA. More water in the system creates more pores, leading to lower strength.
  4. 4) The content of coarse and fine aggregates has a minor impact on strength unless the amount of aggregate is excessive, preventing the paste from completely enveloping the aggregate and resulting in a porous concrete with significantly reduced strength.
  5. 5) Increasing the curing temperature enhances the compressive strength as it accelerates the reaction rate.
  6. 6) At high temperatures, e.g., 80°C, extending the curing time only marginally improves the compressive strength, resulting from the high reaction rate at high temperatures has made concrete to mature in a short period of time.

In summary, Na/Al, H2O/FA, and T are identified as key factors that obviously affect the compressive strength of FAGC.

thumbnail
Download:
Fig 7. Effects of Na/Al, Si/Al, H2O/FA, C/FA, F/FA, T, t1, and t2 on compressive strength.

https://doi.org/10.1371/journal.pone.0310422.g007

3.3. Program for optimal mixture design

The optimal mixture design method that satisfies target compressive strength and minimum CO2 emissions is developed based on the ensemble strength model and LCA-based CO2 emissions model. It is an optimization problem. The mixture and curing parameters x = (MCoarseAggregate, MFineAggregate, MFlyAsh, MWaterglass, MNaOHSolution, PNaOH,NS, T, t1), are the variables to be determined. These parameters represent the mass of each raw material, as well as the heating-curing temperature and time. The objective function is to minimize CO2 emissions, and achieving the desired compressive strength is one main constraint. Based on the objective function and constraints, the optimal mixture design program is implemented using the scipy.optimize.minimize algorithm within Python’s optimization toolbox. To facilitate the input of parameters, this study provides a graphical user interface (GUI), as depicted in Fig 8.

thumbnail
Download:
Fig 8. GUI of the program of the optimal FAGC mixture design.

https://doi.org/10.1371/journal.pone.0310422.g008

Once the user provides the necessary inputs, such as the parameters related to raw materials, curing, and transport, the required H2O/FA, and the desired target compressive strength, the program runs the optimization algorithm to find the optimal FAGC formula that minimizes CO2 emissions while satisfying the specified constraints.

The program outputs the optimal FAGC formula, including the mass fractions of each raw material and the curing parameters. Additionally, it calculates the carbon emissions associated with the optimal mixture design. Furthermore, the program provides information on the reduction in carbon emissions compared to an equivalent cement concrete with the same compressive strength and aggregate volume. The mixture design method of cement concrete follows the Chinese standard JGJ 55–2011 [145], where P·O 42.5 cement is used. P·O refers to ordinary Portland cement (OPC) that contains 80%-95% clinker and 5%-20% other materials such as slag or fly ash by mass, while the number 42.5 represents the strength class.

3.4. Case study for mixture design

Two cases about mixture design are presented, namely one using the optimal mixture design method developed in this study, and the other using the traditional trial and error design method. The latter involves creating multiple trial groups by adjusting the mixture and curing parameters (such as Na/Al and T) and testing them to find the one that achieves the target performance. The target compressive strength for both cases is set to 50 MPa, and the required parameters are shown in Fig 8.

Table 7 shows the design results. Both two mixtures can achieve the target strength. The traditional trial and error design result is from the experimental results as shown in S4 Table. Its CO2 emissions are 160.86 kgCO2/m3, which is calculated using Eq (13), while the carbon emissions of the FAGC mixture designed using the optimal mixture design method are only 134.08 kgCO2/m3, namely a decrease of 16.65%.

thumbnail
Download:
Table 7. Design results of FAGC.

https://doi.org/10.1371/journal.pone.0310422.t007

Comparing the two design methods, the optimal mixture design method exhibits lower water glass content and shorter heat-curing time but a higher heat-curing temperature compared to the trial and error method. This is consistent with the explanation of the model in Section 3.2. It is observed that when the Si content is already high, further increasing it has little effect on strength improvement. Additionally, at high temperatures, extending the curing time only marginally improves the compressive strength, while increasing the heat-curing temperature significantly enhances the compressive strength. The optimization algorithm chooses to reduce the water glass content and heat-curing time while increasing the heat-curing temperature to effectively compensate for the strength loss in a cost-effective manner.

Furthermore, this study compares the CO2 emissions of the optimal FAGC with equivalent cement concrete having the same compressive strength and aggregate volume. The CO2 emissions of the cement concrete are 337.29 kgCO2/m3, while the optimal FAGC emits only 134.08 kgCO2/m3, achieving a significant reduction of 60.25%. This highlights the potential of FAGC as a low-carbon alternative to cement concrete.

4. Conclusions

This study focuses on computational mixture design optimization to develop FAGC that meets key performance targets, such as compressive strength, while minimizing CO2 emissions. First, a robust model for predicting the compressive strength of FAGC is developed. Comprehensive factors, including fly ash characteristics, mixture proportions, curing parameters, and specimen types, are considered. A big dataset comprising 1136 observations is established. Polynomial regression, genetic programming, and ensemble learning techniques are employed. Subsequently, using the developed strength model and a life cycle assessment (LCA)-based CO2 emissions model, an optimal FAGC mixture design program is formulated. The key findings are as follows:

  • The ensemble learning model exhibits superior accuracy and generalization ability compared to single models for compressive strength prediction of FAGC, achieving the lowest RMSE value of 1.91 MPa, the highest R2 value of 0.93, and the highest a20-index value of 1.
  • The experimental data and the ensemble learning model identify Na to Al molar ratio (Na/Al), water to fly ash mass ratio (H2O/FA), and curing temperature (T) as the key factors influencing the compressive strength of FAGC.
  • In the FAGC mixture design case, the optimal design, as opposed to traditional trial-and-error methods, leads to a 16.7% reduction in CO2 emissions for FAGC with a compressive strength of 50 MPa. Moreover, compared to conventional cement concrete, the developed FAGC demonstrates a substantial 61% reduction in CO2 emissions.
  • The results of this work provide engineers with tools for compressive strength prediction and low-carbon optimization of FAGC, enabling rapid and highly accurate design of concrete with lower CO2 emissions and greater sustainability.

5. Limitations and future work

In modeling the compressive strength of FAGC, this study successfully employs an ensemble machine learning model that demonstrates superior performance compared to single models. However, there is room for improvement in the proposed approach. Future enhancements could include integrating an artificial neural network as a third base learner in the ensemble, alongside polynomial regression and genetic programming models, and using genetic programming or artificial neural network instead of linear regression to aggregate base learners and create the ensemble model. In addition, although this work considers a more comprehensive set of input variables than previous work, some influential factors remain unaccounted for, particularly fly ash characteristics such as amorphous content, fineness, and density. This is because most studies do not report this information. Future experiments that provide this information could lead to the inclusion of additional input parameters, thereby improving prediction accuracy.

Regarding mixture design optimization, this study focuses on developing FAGC that meets target compressive strength while minimizing CO2 emissions. Building on the ensemble modeling method proposed in this work, and using the database of workability and durability, future work could develop prediction models for these additional performance metrics. These models could then be integrated into the optimization framework, enabling multi-performance optimization, e.g., meeting target compressive strength, workability, and durability while minimizing CO2 emissions.

References

  1. 1. Monteiro PJ, Miller SA, Horvath A. Towards sustainable concrete. Nat Mater. 2017;16(7):698–699. pmid:28653697
  2. 2. Habert G, Miller SA, John VM, Provis JL, Favier A, Horvath A, et al. Environmental impacts and decarbonization strategies in the cement and concrete industries. Nat Rev Earth Environ. 2020;1(11):559–573.
  3. 3. Giergiczny Z. Fly ash and slag. Cem Concr Res. 2019;124:105826.
  4. 4. Luan C, Zhou A, Li Y, Zou D, Gao P, Liu T. CO2 avoidance cost of fly ash geopolymer concrete. Constr Build Mater. 2024;416:135193.
  5. 5. McLellan BC, Williams RP, Lay J, Van Riessen A, Corder GD. Costs and carbon emissions for geopolymer pastes in comparison to ordinary portland cement. J Clean Prod. 2011;19(9–10):1080–1090.
  6. 6. Fernandez-Jimenez AM, Palomo A, Lopez-Hombrados C. Engineering properties of alkali-activated fly ash concrete. ACI Mater J. 2006;103(2):106.
  7. 7. Luan C, Wang Q, Yang F, Zhang K, Utashev N, Dai J, et al. Practical prediction models of tensile strength and reinforcement-concrete bond strength of low-calcium fly ash geopolymer concrete. Polymers (Basel). 2021;13(6):875. pmid:33809247
  8. 8. Bakharev T. Durability of geopolymer materials in sodium and magnesium sulfate solutions. Cem Concr Res. 2005;35(6):1233–1246.
  9. 9. Kong DL, Sanjayan JG, Sagoe-Crentsil K. Comparative performance of geopolymers made with metakaolin and fly ash after exposure to elevated temperatures. Cem Concr Res. 2007;37(12):1583–1589.
  10. 10. Shill SK, Al-Deen S, Ashraf M, Hutchison W. Resistance of fly ash based geopolymer mortar to both chemicals and high thermal cycles simultaneously. Constr Build Mater. 2020;239:117886.
  11. 11. DeRousseau MA, Kasprzyk JR, Srubar WV III. Computational design optimization of concrete mixtures: A review. Cem Concr Res. 2018;109:42–53.
  12. 12. Young BA, Hall A, Pilon L, Gupta P, Sant G. Can the compressive strength of concrete be estimated from knowledge of the mixture proportions? New insights from statistical analysis and machine learning methods. Cem Concr Res. 2019;115:379–388.
  13. 13. A concrete future. Commun Eng. 2022;1:32.
  14. 14. Li Z, Yoon J, Zhang R, Rajabipour F, Srubar WV III, Dabo I, et al. Machine learning in concrete science: Applications, challenges, and best practices. Npj Comput Mater. 2022;8(1):1–17.
  15. 15. Kumar M, Samui P, Kumar DR, Asteris PG. State-of-the-art XGBoost, RF and DNN based soft-computing models for PGPN piles. Geomech Geoeng. 2024;1–16.
  16. 16. Asteris PG, Lemonis ME, Le TT, Tsavdaridis KD. Evaluation of the ultimate eccentric load of rectangular CFSTs using advanced neural network modeling. Eng Struct. 2021;248:113297.
  17. 17. Asteris PG, Tsaris AK, Cavaleri L, Repapis CC, Papalou A, Di Trapani F, et al. Prediction of the fundamental period of infilled RC frame structures using artificial neural networks. Comput Intell Neurosci. 2016;2016:5104907. pmid:27066069
  18. 18. Zhang F, Wang C, Liu J, Zou X, Sneed LH, Bao Y, et al. Prediction of FRP-concrete interfacial bond strength based on machine learning. Eng Struct. 2023;274:115156.
  19. 19. Pandey S, Paudel S, Devkota K, Kshetri K, Asteris PG. Machine learning unveils the complex nonlinearity of concrete materials’ uniaxial compressive strength. Int J Constr Manag. 2024;1–15.
  20. 20. Alkayem NF, Shen L, Mayya A, Asteris PG, Fu R, Di Luzio G, et al. Prediction of concrete and FRC properties at high temperature using machine and deep learning: a review of recent advances and future perspectives. J Build Eng. 2023;108369.
  21. 21. Armaghani DJ, Asteris PG. A comparative study of ANN and ANFIS models for the prediction of cement-based mortar materials compressive strength. Neural Comput Appl. 2021;33:4501–4532.
  22. 22. Paruthi S, Khan AH, Kumar A, Kumar F, Hasan MA, Magbool HM, et al. Sustainable cement replacement using waste eggshells: A review on mechanical properties of eggshell concrete and strength prediction using artificial neural network. Case Stud Constr Mater. 2023;18.
  23. 23. Paruthi S, Husain A, Alam P, Khan AH, Hasan MA, Magbool HM. A review on material mix proportion and strength influence parameters of geopolymer concrete: Application of ANN model for GPC strength prediction. Constr Build Mater. 2022;356:129253.
  24. 24. Rathnayaka M, Karunasinghe D, Gunasekara C, Wijesundara K, Lokuge W, Law DW. Machine learning approaches to predict compressive strength of fly ash-based geopolymer concrete: A comprehensive review. Constr Build Mater. 2024;419:135519.
  25. 25. Paruthi S, Rahman I, Husain A. Utilizing ANFIS for strength characteristics forecasting in variable heat-cured geopolymer composites. Mater Today Proc. 2024.
  26. 26. Asteris PG, Skentou AD, Bardhan A, Samui P, Pilakoutas K. Predicting concrete compressive strength using hybrid ensembling of surrogate machine learning models. Cem Concr Res. 2021;145:106449.
  27. 27. Lokuge W, Wilson A, Gunasekara C, Law DW, Setunge S. Design of fly ash geopolymer concrete mix proportions using Multivariate Adaptive Regression Spline model. Constr Build Mater. 2018;166:472–481.
  28. 28. Nguyen KT, Nguyen QD, Le TA, Shin J, Lee K. Analyzing the compressive strength of green fly ash based geopolymer concrete using experiment and machine learning approaches. Constr Build Mater. 2020;247:118581.
  29. 29. Huynh AT, Nguyen QD, Xuan QL, Magee B, Chung T, Tran KT, et al. A machine learning-assisted numerical predictor for compressive strength of geopolymer concrete based on experimental data and sensitivity analysis. Appl Sci. 2020;10(21):7726.
  30. 30. Khan MA, Zafar A, Akbar A, Javed MF, Mosavi A. Application of Gene Expression Programming (GEP) for the prediction of compressive strength of geopolymer concrete. Materials (Basel). 2021;14(5):1106. pmid:33652972
  31. 31. Toufigh V, Jafari A. Developing a comprehensive prediction model for compressive strength of fly ash-based geopolymer concrete (FAGC). Constr Build Mater. 2021;277:122241.
  32. 32. Chu HH, Khan MA, Javed M, Zafar A, Khan MI, Alabduljabbar H, et al. Sustainable use of fly-ash: Use of gene-expression programming (GEP) and multi-expression programming (MEP) for forecasting the compressive strength geopolymer concrete. Ain Shams Eng J. 2021;12(4):3603–3617.
  33. 33. Ahmed HU, Mohammed AS, Mohammed AA, Faraj RH. Systematic multiscale models to predict the compressive strength of fly ash-based geopolymer concrete at various mixture proportions and curing regimes. PLoS One. 2021;16(6):e0253006. pmid:34125869
  34. 34. Peng Y, Unluer C. Analyzing the mechanical performance of fly ash-based geopolymer concrete with different machine learning techniques. Constr Build Mater. 2022;316:125785.
  35. 35. Ahmad A, Ahmad W, Aslam F, Joyklad P. Compressive strength prediction of fly ash-based geopolymer concrete via advanced machine learning techniques. Case Stud Constr Mater. 2022;16:e00840.
  36. 36. Khalaf AA, Kopecskó K, Merta I. Prediction of the compressive strength of fly ash geopolymer concrete by an optimised neural network model. Polymers (Basel). 2022;14(7):1423. pmid:35406295
  37. 37. Li Q, Song Z. Prediction of compressive strength of rice husk ash concrete based on stacking ensemble learning model. J Clean Prod. 2022;382:135279.
  38. 38. Yi ST, Yang EI, Choi JC. Effect of specimen sizes, specimen shapes, and placement directions on compressive strength of concrete. Nucl Eng Des. 2006;236(2):115–127.
  39. 39. Ge X, Hu X, Shi C. The effect of different types of class F fly ashes on the mechanical properties of geopolymers cured at ambient environment. Cem Concr Compos. 2022;130:104528.
  40. 40. Singh NB. Fly ash-based geopolymer binder: A future construction material. Minerals. 2018;8(7):299.
  41. 41. Paruthi S, Rahman I, Husain A, Khan AH, Manea-Saghin A, Sabi E. A comprehensive review of nano materials in geopolymer concrete: Impact on properties and performance. Dev Built Environ. 2023;16:100287.
  42. 42. Kaze RC, à Moungam LMB, Cannio M, Rosa R, Kamseu E, Melo UC, et al. Microstructure and engineering properties of Fe2O3 (FeO)-Al2O3-SiO2 based geopolymer composites. J Clean Prod. 2019;199:849–859.
  43. 43. Hu Y, Liang S, Yang J, Chen Y, Ye N, Ke Y, et al. Role of Fe species in geopolymer synthesized from alkali-thermal pretreated Fe-rich Bayer red mud. Constr Build Mater. 2019;200:398–407.
  44. 44. Puligilla S, Mondal P. Role of slag in microstructural development and hardening of fly ash-slag geopolymer. Cem Concr Res. 2013;43:70–80.
  45. 45. Nath P, Sarker PK. Use of OPC to improve setting and early strength properties of low calcium fly ash geopolymer concrete cured at room temperature. Cem Concr Compos. 2015;55:205–214.
  46. 46. Shi X, Zhang C, Wang X, Zhang T, Wang Q. Response surface methodology for multi-objective optimization of fly ash-GGBS based geopolymer mortar. Constr Build Mater. 2022;315:125644.
  47. 47. Provis JL. Alkali-activated materials. Cem Concr Res. 2018;114:40–48.
  48. 48. Paruthi S, Rahman I, Husain A, Hasan MA, Khan AH. Effects of chemicals exposure on the durability of geopolymer concrete incorporated with silica fumes and nano-sized silica at varying curing temperatures. Materials (Basel). 2022;16(18):6332.
  49. 49. Provis JL, Arbi K, Bernal SA, Bondar D, Buchwald A, Castel A, et al. RILEM TC 247-DTA round robin test: mix design and reproducibility of compressive strength of alkali-activated concretes. Mater Struct. 2019;52:1–13.
  50. 50. Assi LN, Deaver E, ElBatanouny MK, Ziehl P. Investigation of early compressive strength of fly ash-based geopolymer concrete. Constr Build Mater. 2016;112:807–815.
  51. 51. Shaikh FUA, Vimonsatit V. Compressive strength of fly-ash-based geopolymer concrete at elevated temperatures. Fire Mater. 2015;39(2):174–188.
  52. 52. Muthadhi A, Suganya B. Effect of Activator on Strength and Microstructure of Alkali Activated Concrete with Class C Fly Ash. Iran J Sci Technol Trans Civ Eng. 2022;46(1):261–269.
  53. 53. Mehta A, Siddique R, Ozbakkaloglu T, Uddin Ahmed Shaikh F, Belarbi R. Fly ash and ground granulated blast furnace slag-based alkali-activated concrete: Mechanical, transport and microstructural properties. Constr Build Mater. 2020;257:119548.
  54. 54. Mehta A, Siddique R. Sulfuric acid resistance of fly ash based geopolymer concrete. Constr Build Mater. 2017;146:136–143.
  55. 55. Rai B, Roy LB, Rajjak M. A statistical investigation of different parameters influencing compressive strength of fly ash induced geopolymer concrete. Struct Concr. 2018;19(5):1268–1279.
  56. 56. Zhang H, Li L, Sarker PK, Long T, Shi X, Wang Q, et al. Investigating various factors affecting the long-term compressive strength of heat-cured fly ash geopolymer concrete and the use of orthogonal experimental design method. Int J Concr Struct Mater. 2019;13(1):63.
  57. 57. Abdulrahman H, Muhamad R, Visintin P, Azim Shukri A. Mechanical properties and bond stress-slip behaviour of fly ash geopolymer concrete. Constr Build Mater. 2022;327:126909.
  58. 58. Nguyen TT, Goodier IC, Austin SA. Factors affecting the slump and strength development of geopolymer concrete. Constr Build Mater. 2020;261:120024.
  59. 59. Ruengsillapanun K, Udtaranakron T, Pulngern T, Tangchirapat W, Jaturapitakkul C. Mechanical properties, shrinkage, and heat evolution of alkali activated fly ash concrete. Constr Build Mater. 2021;299:123954.
  60. 60. Law DW, Adam AA, Molyneaux TK, Patnaikuni I, Wardhono A. Long term durability properties of class F fly ash geopolymer concrete. Mater Struct. 2015;48(3):721–731.
  61. 61. Gunasekera C, Setunge S, Law DW. Correlations between Mechanical Properties of Low-Calcium Fly Ash Geopolymer Concretes. J Mater Civ Eng. 2017;29(9):04017086.
  62. 62. Gunasekara C, Law DW, Setunge S. Long term permeation properties of different fly ash geopolymer concretes. Constr Build Mater. 2016;124:352–362.
  63. 63. Topark-Ngarm P, Chindaprasirt P, Sata V. Setting time, strength, and bond of high-calcium fly ash geopolymer concrete. J Mater Civ Eng. 2015;27(7):04014205.
  64. 64. Thomas RJ, Peethamparan S. Alkali-activated concrete: Engineering properties and stress–strain behavior. Constr Build Mater. 2015;93:49–56.
  65. 65. Nuaklong P, Sata V, Chindaprasirt P. Influence of recycled aggregate on fly ash geopolymer concrete properties. J Clean Prod. 2016;112(4):2300–2307.
  66. 66. Min FS, Yong HC, Ming LY, Abdullah MMAB, Razi HM, Low FW, et al. Effect of silica fume and alumina addition on the mechanical and microstructure of fly ash geopolymer concrete. Arch Metall Mater. 2022;67(1):197–202.
  67. 67. Farhan NA, Sheikh MN, Hadi MNS. Investigation of engineering properties of normal and high strength fly ash based geopolymer and alkali-activated slag concrete compared to ordinary Portland cement concrete. Constr Build Mater. 2019;196:26–42.
  68. 68. Le TA, Le SH, Nguyen TN, Nguyen KT. Assessment of the rheological and mechanical properties of geopolymer concrete comprising fly ash and fluid catalytic cracking residue as aluminosilicate precursor. Appl Sci. 2021;11(7):3032.
  69. 69. Sarker PK. Bond strength of reinforcing steel embedded in fly ash-based geopolymer concrete. Mater Struct. 2011;44(5):1021–1030.
  70. 70. Deb PS, Nath P, Sarker PK. The effects of ground granulated blast-furnace slag blending with fly ash and activator content on the workability and strength properties of geopolymer concrete cured at ambient temperature. Mater Des (1980–2015). 2014;62:32–39.
  71. 71. Yang W, Zhu P, Liu H, Wang X, Ge W, Hua M. Resistance to sulfuric acid corrosion of geopolymer concrete based on different binding materials and alkali concentrations. Materials (Basel). 2021;14(23):7109. pmid:34885264
  72. 72. Ali IM, Naje AS, Al-Zubaidi HAM, Al-Kateeb RT. Performance evaluation of fly ash-based geopolymer concrete incorporating nano slag. Glob Nest J. 2019;21(1):70–75.
  73. 73. Okoye FN, Durgaprasad J, Singh NB. Effect of silica fume on the mechanical properties of fly ash based-geopolymer concrete. Ceram Int. 2016;42(2):3000–3006.
  74. 74. Nagajothi S, Elavenil S. Effect of GGBS addition on reactivity and microstructure properties of ambient cured fly ash based geopolymer concrete. Silicon. 2021;13(2):507–516.
  75. 75. Mehta A, Siddique R. Strength, permeability and micro-structural characteristics of low-calcium fly ash based geopolymers. Constr Build Mater. 2017;141:325–334.
  76. 76. Mehta A, Siddique R. Properties of low-calcium fly ash based geopolymer concrete incorporating OPC as partial replacement of fly ash. Constr Build Mater. 2017;150:792–807.
  77. 77. Chindaprasirt P, Chalee W. Effect of sodium hydroxide concentration on chloride penetration and steel corrosion of fly ash-based geopolymer concrete under marine site. Constr Build Mater. 2014;63:303–310.
  78. 78. Gunasekara C, Dirgantara R, Law DW, Setunge S. Effect of curing conditions on microstructure and pore-structure of brown coal fly ash geopolymers. Appl Sci. 2019;9(15):3138.
  79. 79. Memon FA, Nuruddin MF, Shafiq N. Effect of silica fume on the fresh and hardened properties of fly ash-based self-compacting geopolymer concrete. Int J Miner Metall Mater. 2013;20(2):205–213.
  80. 80. Gomaa E, Sargon S, Kashosi C, Gheni A, ElGawady MA. Mechanical properties of high early strength class C fly ash-based alkali activated concrete. Transp Res Rec. 2020;2674(5):430–443.
  81. 81. Nuruddin MF, Demie S, Shafiq N. Effect of mix composition on workability and compressive strength of self-compacting geopolymer concrete. Can J Civ Eng. 2011;38(11):1196–1203.
  82. 82. Cui Y, Gao K, Zhang P. Experimental and statistical study on mechanical characteristics of geopolymer concrete. Materials (Basel). 2020;13(7):1651. pmid:32252367
  83. 83. Olivia M, Nikraz H. Properties of fly ash geopolymer concrete designed by Taguchi method. Mater Des (1980–2015). 2012;36:191–198.
  84. 84. Verma M, Dev N. Effect of ground granulated blast furnace slag and fly ash ratio and the curing conditions on the mechanical properties of geopolymer concrete. Struct Concr. 2022;23(4):2015–2029.
  85. 85. Zhang H, Li L, Sarker PK, Long T, Shi X, Cai G, et al. The effect of ordinary Portland cement substitution on the thermal stability of geopolymer concrete. Materials (Basel). 2019;12(16):2501. pmid:31394771
  86. 86. Long T, Wang Q, Guan Z, Chen Y, Shi X. Deterioration and microstructural evolution of the fly ash geopolymer concrete against MgSO4 solution. Adv Mater Sci Eng. 2017;2017:1–11.
  87. 87. Negahban E, Bagheri A, Sanjayan J. Pore gradation effect on Portland cement and geopolymer concretes. Cem Concr Compos. 2021;122:104141.
  88. 88. Diaz-Loya EI, Allouche EN, Vaidya S. Mechanical properties of fly-ash-based geopolymer concrete. ACI Mater J. 2011;108(3):300–306.
  89. 89. Cornelis R, Priyosulityo H, Satyarno I, Rustendi I. Effect of the mortar volume ratio on the mechanical behavior of class CI fly ash-based geopolymer concrete. Civ Eng J. 2022;8(9):1920–1935.
  90. 90. Wardhono A, Law DW, Molyneaux TK. The mechanical properties of fly ash geopolymer in long term performance. In: CIC2014 Concrete Innovation Conference, Oslo, Norway. 2014.
    • 91. Shehab HK, Eisa AS, Wahba AM. Mechanical properties of fly ash based geopolymer concrete with full and partial cement replacement. Constr Build Mater. 2016;126:560–565.
    • 92. Hongen Z, Feng J, Qingyuan W, Ling T, Xiaoshuang S. Influence of cement on properties of fly-ash-based concrete. ACI Mater J. 2017;114(5):745–753.
    • 93. Pachamuthu S, Thangaraju P. Effect of incinerated paper sludge ash on fly ash-based geopolymer concrete. J Croatian Assoc Civ Eng. 2017;69(9):851–859.
    • 94. Vora PR, Dave UV. Parametric studies on compressive strength of geopolymer concrete. Procedia Eng. 2013;51:210–219.
    • 95. Zhang H, Li L, Yuan C, Wang Q, Sarker PK, Shi X. Deterioration of ambient-cured and heat-cured fly ash geopolymer concrete by high temperature exposure and prediction of its residual compressive strength. Constr Build Mater. 2020;262:120924.
    • 96. Ahmed MF, Nuruddin MF, Shafiq N. Compressive strength and workability characteristics of low-calcium fly ash-based self-compacting geopolymer concrete. Int J Civ Environ Eng. 2011;5(2):64–70.
    • 97. Hassan A, Arif M, Shariq M. Effect of curing condition on the mechanical properties of fly ash-based geopolymer concrete. SN Appl Sci. 2019;1(12):1–9.
    • 98. Memon FA, Nuruddin MF, Demie S, Shafiq N. Effect of curing conditions on strength of fly ash-based self-compacting geopolymer concrete. Int J Civ Environ Eng. 2011;5(8):342–345.
    • 99. Wallah SE. Creep behaviour of fly ash-based geopolymer concrete. Civ Eng Dimens. 2010;12(2):73–78.
    • 100. Sarker PK, Haque R, Ramgolam KV. Fracture behaviour of heat cured fly ash based geopolymer concrete. Mater Des (1980–2015). 2013;44:580–586.
    • 101. Shaikh FUA. Mechanical and durability properties of fly ash geopolymer concrete containing recycled coarse aggregates. Int J Sustain Built Environ. 2016;5(2):277–287.
    • 102. Patankar SV, Jamkar SS, Ghugal YM. Effect of water-to-geopolymer binder ratio on the production of fly ash based geopolymer concrete. Int J Adv Technol Civ Eng. 2013;2(1):79–83.
    • 103. Olivia M, Nikraz H. Strength and water penetrability of fly ash geopolymer concrete. J Eng Appl Sci. 2011;6(7):70–78.
    • 104. Olivia M, Nikraz H. Water penetrability of low calcium fly ash geopolymer concrete. https://repository.unri.ac.id/bitstream/handle/123456789/3577/Monita%20Olivia_ICCBT_2008.pdf?sequence=3&isAllowed=y.
      • 105. Gunasekara C, Atzarakis P, Lokuge W, Law DW, Setunge S. Novel analytical method for mix design and performance prediction of high calcium fly ash geopolymer concrete. Polymers (Basel). 2021;13(6):900. pmid:33804194
      • 106. Demie S, Nuruddin MF, Shafiq N. Effects of micro-structure characteristics of interfacial transition zone on the compressive strength of self-compacting geopolymer concrete. Constr Build Mater. 2013;41:91–98.
      • 107. Çevik A, Alzeebaree R, Humur G, Niş A, Gülşan ME. Effect of nano-silica on the chemical durability and mechanical performance of fly ash based geopolymer concrete. Ceram Int. 2018;44(11):12253–12264.
      • 108. Okoye FN, Durgaprasad J, Singh NB. Mechanical properties of alkali activated flyash/Kaolin based geopolymer concrete. Constr Build Mater. 2015;98:685–691.
      • 109. Jena S, Panigrahi R, Sahu P. Mechanical and durability properties of fly ash geopolymer concrete with silica fume. J Inst Eng India Ser A. 2019;100(4):697–705.
      • 110. Wang Y, Hu S, He Z. Mechanical and fracture properties of fly ash geopolymer concrete additive with calcium aluminate cement. Materials (Basel). 2019;12(18):2982.
      • 111. Bellum RR, Venkatesh C, Madduru SRC. Influence of red mud on performance enhancement of fly ash-based geopolymer concrete. Innov Infrastruct Solut. 2021;6(4):1–9.
      • 112. Embong R, Kusbiantoro A, Shafiq N, Nuruddin MF. Strength and microstructural properties of fly ash based geopolymer concrete containing high-calcium and water-absorptive aggregate. J Clean Prod. 2016;112:816–822.
      • 113. Pan Z, Sanjayan JG, Rangan BV. Fracture properties of geopolymer paste and concrete. Mag Concr Res. 2011;63(10):763–771.
      • 114. Kusbiantoro A, Nuruddin MF, Shafiq N, Qazi SA. The effect of microwave incinerated rice husk ash on the compressive and bond strength of fly ash based geopolymer concrete. Constr Build Mater. 2012;36:695–703.
      • 115. Joseph B, Mathew G. Influence of aggregate content on the behavior of fly ash based geopolymer concrete. Sci Iran. 2012;19(5):1188–1194.
      • 116. Vikas G, Rao TDG. Setting time, workability and strength properties of alkali activated fly ash and slag based geopolymer concrete activated with high silica modulus water glass. Iran J Sci Technol Trans Civ Eng. 2021;45(3):1483–1492.
      • 117. Haruna S, Mohammed BS, Wahab MMA, Liew MS. Effect of paste aggregate ratio and curing methods on the performance of one-part alkali-activated concrete. Constr Build Mater. 2020;261:120024.
      • 118. Kar A, Ray I, Unnikrishnan A, Halabe UB. Prediction models for compressive strength of concrete with alkali-activated binders. Comput Concr. 2016;17(4):523–539.
      • 119. Aleem MA, Arumairaj PD. Optimum mix for the geopolymer concrete. Indian J Sci Technol. 2012;5(3):2299–2301.
      • 120. Wardhono A, Gunasekara C, Law DW, Setunge S. Comparison of long term performance between alkali activated slag and fly ash geopolymer concretes. Constr Build Mater. 2017;143:272–279.
      • 121. Luan C, Shi X, Zhang K, Utashev N, Yang F, Dai J, et al. A mix design method of fly ash geopolymer concrete based on factors analysis. Constr Build Mater. 2021;272:121612.
      • 122. Shekhovtsova J, Kovtun M, Kearsley EP. Evaluation of short- and long-term properties of heat-cured alkali-activated fly ash concrete. Mag Concr Res. 2015;67(16):897–905.
      • 123. Mortar NAM, Kamarudin H, Rafiza RA, Meor TAF, Rosnita M. Compressive strength of fly ash geopolymer concrete by varying sodium hydroxide molarity and aggregate to binder ratio. IOP Conf Ser Mater Sci Eng. 2020;864(1):012037.
      • 124. Xie T, Ozbakkaloglu T. Behavior of low-calcium fly and bottom ash-based geopolymer concrete cured at ambient temperature. Ceram Int. 2015;41(4):5945–5958.
      • 125. Bocullo V, Vaiciukyniene D, Gecys R, Dauksys M. Effect of ordinary Portland cement and water glass on the properties of alkali activated fly ash concrete. Minerals. 2020;10(1):21.
      • 126. Chithambaram SJ, Kumar S, Prasad MM, Adak D. Effect of parameters on the compressive strength of fly ash based geopolymer concrete. Struct Concr. 2018;19(4):1202–1209.
      • 127. Al-Azzawi M, Yu T, Hadi MNS. Factors affecting the bond strength between the fly ash-based geopolymer concrete and steel reinforcement. Structures. 2018;15:203–215.
      • 128. Hardjito D, Rangan BV. Development and properties of low-calcium fly ash-based geopolymer concrete. https://espace.curtin.edu.au/bitstream/handle/20.500.11937/5594/19327_downloaded_stream_419.pdf?sequence=2.
        • 129. Dai J, Shi X, Wang Q, Zhang H, Luan C, Zhang K, et al. Effect of multi-factor on the compressive strength of construction and demolition waste based geopolymer concrete. Mater Rep. 2021;35(9):9077–9082. (in Chinese).
        • 130. GPlearn documentation. https://gplearn.readthedocs.io/en/stable/index.html.
          • 131. Standard for test methods of concrete physical and mechanical properties. Chinese standard GB/T 50081–2019. (in Chinese).
            • 132. Testing hardened concrete—Part 3: Compressive strength of test specimens. European standard EN 12390–3:2009.
              • 133. Apostolopoulou M, Asteris PG, Armaghani DJ, Douvika MG, Lourenço PB, Cavaleri L, et al. Mapping and holistic design of natural hydraulic lime mortars. Cem Concr Res. 2020;136:106167.
              • 134. Asteris PG, Koopialipoor M, Armaghani DJ, et al. Prediction of cement-based mortars compressive strength using machine learning techniques. Neural Comput Appl. 2021;33:13089–13121.
              • 135. Asteris PG, Lourenço PB, Hajihassani M, Adami CN, Lemonis ME, Skentou AD, et al. Soft computing-based models for the prediction of masonry compressive strength. Eng Struct. 2021;248:113276.
              • 136. Habert G, Ouellet-Plamondon C. Recent update on the environmental impact of geopolymers. RILEM Tech Lett. 2016;1:17–23.
              • 137. Shobeiri V, Bennett B, Xie T, Visintin P. A comprehensive assessment of the global warming potential of geopolymer concrete. J Clean Prod. 2021;297:126669.
              • 138. Turner LK, Collins FG. Carbon dioxide equivalent (CO2-e) emissions: A comparison between geopolymer and OPC cement concrete. Constr Build Mater. 2013;43:125–130.
              • 139. Salas DA, Ramirez AD, Ulloa N, Baykara H, Boero AJ. Life cycle assessment of geopolymer concrete. Constr Build Mater. 2018;190:170–177.
              • 140. Chu S. Effect of paste volume on fresh and hardened properties of concrete. Constr Build Mater. 2019;218:284–294.
              • 141. Rees CA, Provis JL, Lukey GC, Van Deventer JS. In situ ATR-FTIR study of the early stages of fly ash geopolymer gel formation. Langmuir. 2007;23(17):9076–9082. pmid:17658864
              • 142. Wang Y, Liu X, Zhang W, Li Z, Zhang Y, Li Y, et al. Effects of Si/Al ratio on the efflorescence and properties of fly ash based geopolymer. J Clean Prod. 2020;244:118852.
              • 143. Liu J, Doh JH, Dinh HL, Ong DE, Zi G, You I. Effect of Si/Al molar ratio on the strength behavior of geopolymer derived from various industrial waste: A current state of the art review. Constr Build Mater. 2022;329:127134.
              • 144. Li J, Tao Y, Zhuang E, Cui X, Yu K, Yu B, et al. Optimal amorphous oxide ratios and multifactor models for binary geopolymers from metakaolin blended with substantial sugarcane bagasse ash. J Clean Prod. 2022;377:134215.
              • 145. Specification for mix proportion design of ordinary concrete. Chinese standard JGJ 55–2011. (in Chinese).