Journal of Building Engineering 72 (2023) 106611

Contents lists available at ScienceDirect

Journal of Building Engineering

journal homepage: www.elsevier.com/locate/jobe

Experimental optimization of industrial waste-based soil

hardening agent: Combining D-optimal design with
genetic algorithm
Xin Chen, Feng Yu *, Jing Yu, Shuaikang Li
Institute of Foundation and Structure Technologies, Zhejiang Sci-Tech University, Xiasha Higher Education Park, Hangzhou, 310018, PR China

A R T I C L E I N F O A B S T R A C T

Keywords: In this study, an innovative approach was proposed for developing industrial-waste-based soil
Experimental optimization curing agent mixtures using D-optimal design (DOD) and the genetic algorithm (GA). The in­
Soil hardening agent fluences of blast furnace slag (BFS), steel slag (SS), phosphogypsum (PG), and 42.5-grade ordinary
Unconfined compressive strength Portland cement (P.O 42.5) on the mechanical characteristics of the solidified soil were assessed
Genetic algorithm through unconfined compressive strength (UCS) tests. The mechanism, hydrates, and ecological
Life cycle assessment
properties of the soil solidified with the optimized curing agent were also evaluated. The statis­
tical analysis confirmed the precision of the fitted regression model in forecasting the mechanical
characteristics of the stabilized soil. Furthermore, the BFS and PG were found to have a positive
effect on the UCS within a certain range at a 7-day age, while the SS and P.O 42.5 exhibited the
opposite effect. The hybrid DOD-GA technique was more effective to optimize the formulation
than the DOD technique. Overall, the UCS of the new curing agent-stabilized soil cured for 7 days
with the same dosage was 46.25% higher than that of cement-stabilized soil; and the global
warming potential (GWP) of the new curing agent per ton was 70.42% lower than that of cement.
The new curing agent is more capable of achieving improved strength, microstructure, and
environmental benefits than cement is, in addition to having inherent potential in industrial
waste recycling, greenhouse buffering, and natural resource conservation. The contribution of
this research is that the proposed systematic framework is universal, and it is reasonable to
implement formulation optimization of soil hardening agents.

1. Introduction
Global warming continues to intensify due to the enormous emissions of greenhouse gases. According to the International Energy
Agency publication “Global Energy Review: Carbon Dioxide Emissions in 2021,” the global CO2 emissions of the energy field in 2021
totaled 36.3 billion tons [1]. Initial statistics suggest that 2.537 Gt of CO2 in the cement industry was produced globally in 2020 [2].
Further, treatment of industrial waste has become a major problem in the global environment and energy fields, and its utilization
rate is less than 60%. The recycling of industrial waste is essential for achieving the Net Zero Emissions by 2050 Scenario goal. The
development of civil engineering promotes cement demand, which further exacerbates environmental pollution [3]. Therefore, the
cement industry is an important determinant of CO2 emissions and energy consumption [4]. Moreover, cemented soil exhibits the
disadvantage of a low early strength. It is possible to replace cement with industrial waste as the soil curing agent. It has been reported

* Corresponding author.
E-mail addresses: [email protected] (X. Chen), [email protected] (F. Yu), [email protected] (J. Yu), [email protected] (S. Li).

https://doi.org/10.1016/j.jobe.2023.106611
Received 20 February 2023; Received in revised form 1 April 2023; Accepted 18 April 2023
Available online 19 April 2023
2352-7102/© 2023 Elsevier Ltd. All rights reserved.
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

that the appropriate utilization of industrial waste benefits the strength of solidified soil but additionally effectively decreases the
environmental burden [5,6]. For example, Miraki et al. [7] adopted mechanical and microstructural tests to evaluate the durability of
stabilized clay by volcanic ash (VA) and blast furnace slag (BFS). The testing results demonstrated that a proper aggregation of VA and
BFS was superior to Portland cement in terms of the structure, strength, and environment. Liu et al. [8] conducted a series of
experimental tests on the properties of SCM (steel slag, cement, and metakaolin)-solidified soil, which included unconfined
compressive strength (UCS) tests and scanning electron microscopy (SEM). The experimental results showed that the strength of the
SCM-stabilized clay reached 1.0 MPa at the curing age of 28 days, and the relationship between the strength and binder-free water ratio
was proposed for further analysis. Jiang et al. [9] studied a binder composed of BFS, carbide slag (CS), and other materials to stabilize
soft soil instead of cement. A giant quantity of hydrate in the new binder-solidified soil supplied a greater sulfate attack resistance. Li
et al. [10] studied the effect of BFS and cement for as-contaminated soil treatment, which confirmed that the BFS had a prominent
improvement effect on the strength at 28 days. Salimi and Ghorbani [11] carried out several experiments to investigate the various
factors affecting the properties of slag- and geopolymer-stabilized soft clay. Ren et al. [12] examined the effect of
phosphogypsum-solidified silty soil, aiming to establish a linear model to analyze the association between the physical parameters and
the microstructural properties.
The investigations in the available scientific literature have revealed that industrial waste has been considered as an alternative soil
curing agent to cement with improved environmental sustainability [13]. In most of these studies, the optimization of the curing agents
was carried out by varying the variables one by one based on traditional control variable methods, resulting in problems of a high
number of experimental trials and a relatively long period test. However, the interactions between variables are difficult to take into
consideration in univariate approaches. Hence, mixture design is a specially constrained regression design that considers the inter­
action of variables, requiring a lower number of experiments [14]. Mixture design has been applied in many fields, including civil
engineering, since it was first proposed by Scheffé in 1958 [15]. For instance, Wang et al. [16] established a quadratic mixture model to
evaluate the effects of solid particles and superplasticizers on the bulk density, confirming that the D-optimal mixture design was able
to optimize the ultra-high performance concrete (UHPC) bulk density effectively. Varanda et al. [17] conducted a constrained mixture
design to prepare bitumen blends composed of residues and aromatic extracts. Sun et al. [18] added limestone and calcined clay
tailings into cement to optimize the system of the eco-efficient UHPC through a D-optimal mixture design. The eco-efficient UHPC was
superior to the normal UHPC in the environment. However, it had a lower strength and workability. In the aforementioned studies,
mixture design was an effective and essential tool for cementitious material formulation involving numerous components. Mixture
design mainly includes experimental design, model development, model suitability evaluation and optimization. The optimization of
nonlinear model also has difficulty in selecting the optimal parameters. Hence, the metaheuristic algorithm is proposed to effectively
improve the prediction accuracy of regression model. Genetic algorithm (GA) is one of the most widely used metaheuristic algorithms
and well accepted due to its outstanding search capability, randomness as well as good compatibility [19]. Therefore, the GA is
employed in this study for formula optimization.
In this study, industrial wastes, including blast furnace slag (BFS), steel slag (SS), and phosphogypsum (PG), were attempted to be
recycled into solidified soil. More specifically, the UCS at the early stage was treated as the response to assess the mechanical properties
of the stabilized soil, and the D-optimal design (DOD) model was established. In view of the discussion above, the optimal dosages of
BFS, SS, PG, 42.5-grade ordinary Portland cement (P.O 42.5), and superplasticizer (SP) were obtained by establishing a genetic al­
gorithm (GA) model with specific constraints. Then, the strengths of the cured soils obtained by the DOD and DOD-GA methods were
verified. Finally, the curing mechanisms and hydration characteristics of the optimal stabilized soil and cemented soil were studied.
The life cycle assessment (LCA) of the industrial-waste-based soil hardening agent was carried out by the economic value allocation
principle. The contribution of this research is that the proposed systematic framework is universal and reasonable to implement for the
formulation optimization of soil hardening agents, and it provides a reference for similar analyses.

2. Materials and methodology

2.1. Materials
2.1.1. Soil sample
The tested soil sample was silt typically distributed in Hangzhou, China. The physical characteristics of the silt are shown in Table 1.
The original silt was open-air dried, pulverized, and passed through a 2-mm sieve for testing. After the above work, the optimal water
content was determined to be 19.66% through the compaction test (in Fig. 1). Thus, the treated dry silt was mixed with water of
optimum moisture content to prepare the wet silt. The mineral composition of the silt is shown in Fig. 2.

2.1.2. Stabilizers
In this study, BFS, SS, PG, P.O 42.5, and SP were used as the raw materials for generating the new soil hardening agent. The SP was
the polycarboxylate type. The chemical ingredients of the inorganic materials in this study are displayed in Table 2.

Table 1
Physical characteristics of silt.

Soil Unit Weight, Moisture Void Cohesion, Internal Friction Liquid Plastic Compression
Sample kN•m− 3 Content,% Ratio kPa Angle, ◦ Limit,% Limit,% Modulus, MPa

Silt 15.8 14 1.05 6 17 28.41 16.69 3.6

2
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 1. The result of compaction test.

Fig. 2. Mineral composition of silt.

2.2. Experimental design

Considering that the new soil hardening agent was composed of various components, the mixture design was adopted in this paper.
Since it was first proposed by Scheffé in 1958s [15], mixture design has been widely used in the formulation tests in various fields, such
as medicine and food.
A mixture design is a specially constrained regression design consisting of multiple factors, where the sum of the factors is 1. The
response of the mixture design is only related to the content of each component. Therefore, the proportion xi (i = 1, 2...q) of the i-th
ingredient in the q-component mixture system must obey the constraints
⎧∑ q
⎨ xi = x1 + x2 + ... + xq = 1
(1)
⎩ i=1
xi ≥ 0

Table 2
Chemical compositions of inorganic materials in this study.

Materials CaO SiO2 Al2O3 SO3 Fe2O3 MgO

BFS 34.00 34.50 17.70 1.64 4.03 6.01

SS 41.74 16.75 3.87 0.18 20.12 7.35
PG 30.10 8.73 1.23 44.25 0.52 –
P.O 42.5 51.42 24.99 8.26 2.51 4.03 3.71

3
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Table 3
Constrained variables and responses.

Variables Restrictions Response

Actual Coded Min Max Actual Coded

BFS X1(A) 40 60 Unconfined compressive strength y

SS X2(B) 0 15
PG X3(C) 0 15
P.O 42.5 X4(D) 15 30
SP X5(E) 0.1

where xi is the mixture amount.

Due to the limitations of practical problems, other constraints will be added to the mixing system based on Eq. (1). The component
varies in a subinterval within [0, 1], and the mixture problem with upper and lower bound constraints is obtained:
⎧∑ q
⎨ xi = x1 + x2 + ... + xq = 1
(2)
⎩ i=1
0 ≤ ai ≤ xi ≤ bi ≤ 1

where ai and bi represent the lower and upper bounds of xi , respectively.

The factor space of the q-component mixture problem with both upper and lower bound constraints is an irregular convex poly­
hedron in the normal simplex of (q− 1) dimensions. The response changes as the ratios of the components in the mixture change. The
optimal combination can be determined through the functional relationship between the experimental response and the ratio of each
factor based on the response surface method.
It is assumed that
( )
y = φ x1 , x2 , ..., xq (3)
( )
where y represents the response, and φ x1 , x2 , ..., xq is a continuous function of xi (i = 1, 2,...,q). In general, a Taylor series can be used
( )
to expand φ x1 , x2 , ..., xq into a power series to approximate the response function under the assumption of continuity.
The quadratic regression model of the q-components is expressed as follows:
q
∑ q− 1 ∑
∑ q− 1
y= β i xi + βij xi xj (4)
i=1 i=1 j=i+1

where βi and βij are regression coefficients.

Fig. 3. Framework for the optimization of the soil hardening agent.

4
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

There are many approaches for mixture design, including Simplex Centroid, Simplex Lattice, Extreme Vertices, ABCD Design,
Space-Filling, and Optimal designs [20–22]. In this study, the proportions of the components were restricted through higher and lower
limits. Hence, D-optimal design was applicable [23]. D-optimal design is an iterative algorithm that maximizes the determinant of the
information matrix [24]. This method provides an effective means for formula optimization with fewer tests and high precision of the
prediction model fitting.
To evaluate the optimal proportions of BFS, SS, PG, P.O 42.5, and SP, a set of mixture experiments designed by D-optimal design
were carried out. Design-Expert 12.0 was employed for the experiment design, model development, analysis of variance (ANOVA),
diagnostics, and optimization. The design and data analysis of mixture experiments usually include the following steps (Fig. 3):
(i) Determine response and mixture components.
(ii) Select an appropriate design algorithm to solve the problem. In this investigation, the mixing proportions of BFS, SS, PG, P.O
42.5, and SP were determined based on D-optimal design, and the constrained variables and responses are introduced in
Table 3.
(iii) Obtain the matrix of the D-optimal design. A 20-run-point mixture design was generated for the five-component system by the
Scheffé quadratic model, including 10 runs for model fitting, five replicate runs, and five runs for lack of fit evaluation.
(iv) Perform experiments and collect data.
(v) Analyze and verify the fitted model adequacy to determine the correlation between the requirements and factors.
(vi) Determine the optimal mixture components of the new soil hardening agents.

2.3. Methodology
2.3.1. Specimen preparation
The specimen was prepared in accordance with the requirements in “Specification for mix proportion design of cement soil (JGJ/T
233–2011)” [25]. The curing agent content should not be less than 20% of the wet soil mass in accordance with “Technical specifi­
cation for trench cutting re-mixing deep wall (JGJ∕T 303–2013)” [26]. In this study, the ratio of the curing agent mixture to the wet
soil was set to be 20% with respect of engineering practice. The raw materials BFS, SS, PG, and P.O 42.5 were weighed and mixed
uniformly according to the test groups defined in Table 4. Then, the BFS, SS, PG, P.O 42.5, and soil were poured into the mixer for
stirring at a low speed for 1 min. After the mixer stopped, the SP and water were added to the mixture, and the water–binder (W/B)
ratio was 0.8. The mixer was set to a high speed, and the mixture was stirred for 4 min. The created stabilized soil was poured into three
cubic molds with side lengths of 70.7 mm and vibrated. The specimen was demolded after 24 h, and cured film mulching was per­
formed at room temperature.

2.3.2. Unconfined compressive strength (UCS) test

The WAW-300B universal testing machine was utilized to test the UCS of stabilized soil specimens cured for 7 days, with a load rate
of 0.8 kN/s (Fig. 4(b)). Three samples of each group were tested to avoid the contingency of the experimental data. The samples and
equipment of the UCS are shown in Fig. 4. Simultaneously, a group of cement soil samples was prepared to compare with the me­
chanical properties of the new solidified soil.

2.3.3. Hydration characteristics

The hydration characteristics of the stabilized soil were observed by scanning electron microscopy (SEM, Gemini 500) and X-ray
diffraction (XRD, Thermal ARL X’TRA). The SEM and XRD equipment are shown in Fig. 5. During the scanning electron microscope

Table 4
D-optimal mixture design run points and corresponding responses.

Run X1(A) X2(B) X3(C) X4(D) X5(E) y

1 57.0481 2.24503 10.6069 30.00 0.10 2.93

2 60.00 15.00 0.00 24.90 0.10 2.35
3 53.9208 15.00 7.84126 23.1379 0.10 2.45
4 46.4981 10.245 15.00 28.1569 0.10 2.72
5 49.512 15.00 5.388 30.00 0.10 2.26
6 52.8825 9.13828 9.22619 28.653 0.10 2.49
7 49.512 15.00 15.00 20.388 0.10 3.26
8 40.00 14.90 15.00 30.00 0.10 1.83
9 60.00 9.36158 8.42213 22.1163 0.10 2.46
10 49.512 15.00 15.00 20.388 0.10 3.32
11 60.00 15.00 9.90 15.00 0.10 2.44
12 49.512 15.00 5.388 30.00 0.10 2.12
13 60.00 9.90 15.00 15.00 0.10 3.40
14 60.00 0.00 15.00 24.90 0.10 3.49
15 60.00 0.00 15.00 24.90 0.10 3.61
16 54.9202 8.12407 15.00 21.8558 0.10 3.37
17 51.0497 3.85034 15.00 30.00 0.10 2.56
18 40.00 14.90 15.00 30.00 0.10 1.72
19 60.00 15.00 0.00 24.90 0.10 2.18
20 60.00 5.45043 4.44957 30.00 0.10 2.43

5
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 4. Samples and equipment of unconfined compressive strength (UCS): (a) samples and (b) WAW-300B universal testing machine.

(SEM) test, the moisture in the sample will reduce the speed of equipment to reach the vacuum and the service life of the filament.
Therefore, the testing sample was dried at a temperature of 110◦ C to evaporate the free water and crushed. The naturally broken,
relatively flat soil particles were selected to be gold-plated by the JEC-3000FC to be used in the SEM analysis. After passing through a
0.075-mm sieve, the stabilized soil powder could be used in the XRD analysis. The following parameters were selected: Cu (Kα), 40 kV,
40 mA, 10◦ –80◦ , and 3◦ /min.

2.3.4. Life cycle assessment (LCA) analysis—global warming potential

The large amount of exhaust emissions from cement production has aroused societal concern. Therefore, it is necessary to perform
the LCA of the curing agent production process. LCA has been utilized to estimate the potential environmental effects of a construction
process in engineering throughout its life cycles qualitatively and quantitatively [27–29]. LCA consists of four phases: determination of
the purpose and scope, life cycle inventory, life cycle impact assessment, and explanation [30]. A study was proposed to investigate the
potential environmental impact of new soil hardeners. Considering the process from raw material acquisition to the manufactured
product, a “cradle-to-gate” life cycle was selected as the LCA boundary. The specific system boundary is shown in Fig. 6. The global
warming potential (GWP) was used as the LCA index in this paper.

3. Results and discussion

3.1. Mathematical model development
A 20-set test was designed based on the DOD, and the experimental results were analyzed to obtain a quadratic model. The cor­
responding responses of 20 kinds of mixtures at the curing age of 7 days are shown in Table 4 and Fig. 7. It is essential to evaluate the
standard error of the design before model training. Fig. 8 depicts the fraction of design space (FDS) graph, which shows the non-linear
relationship between the standard error mean and the FDS. The standard error (σ ) in the design space was in the range of 0.4–1.1,
which was nearly uniform and small enough to estimate the design precisely.
The two-dimensional (2D) contour plots and three-dimensional (3D) surface plots are shown in Fig. 9. The darker shaded represents
a higher error at a given coordinate in the variable space and vice versa. The standard error was in the range of 0.5 σ–0.7 σ with fixed
contents of P.O 42.5 (D = 25.36 wt%) and SP (E = 0.1 wt%). Therefore, the fitted model could provide exact predictions.
The full quadratic model was developed from the 20 sets of experimental data according to Eq. (4). The significance of the full
quadratic model was determined by the F-test to decide if linear terms, interactions between factors, or secondary terms would be
considered in the prediction model. The significance of the individual and interactive parameters can be identified by the F-value and

Fig. 5. Scanning electron microscopy (SEM) and X-ray diffraction (XRD) equipment: (a) Gemini 500 scanning electron microscope and (b) Thermal ARL X’TRA X-ray
diffractometer.

6
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 6. System boundary of life cycle assessment (LCA) of new soil hardening agent.

Fig. 7. UCSs of 20 kinds of mixtures cured at 7 days.

Fig. 8. Graph of fraction of design space (FDS).

7
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 9. Two-dimensional contour and three-dimensional surface plots with variations of components: (A) blast furnace slag (BFS), (B) steel slag (SS), and (C)
phosphogypsum (PG) with fixed contents of Portland cement (P.O 42.5, D = 25.36 wt%) and superplasticizer (SP, E = 0.1 wt%).

p-value. A high F-value implies that the random error of the model is smaller than the explained variance [31]. This indicates that the
result has statistical significance with a larger F-value and a smaller p-value below 0.05 [32]. For the response of the UCS, the in­
teractions AD, BC, and CD were significant (p < 0.05) and the interactions AB, AC, and BD were insignificant (p > 0.05). Hence, it was
necessary to remove insignificant items in the model through the backward elimination technique (BET) [23]. The BET was imple­
mented by putting all terms into the model simultaneously and then removing one at a time until the best-term-smaller model was
obtained. The acquired final mathematical fitted model based on regression analysis is as follows:
y = − 0.089x1 + 0.158x2 + 0.466x3 − 0.275x4 + 0.008x1 x4 − 0.006x2 x3 − 0.088x3 x4 (5)

where y represents the unconfined compressive strength, x1, x2, x3, and x4 represent the BFS, SS, PG, and P.O 42.5, respectively.

3.2. Analysis of variance (ANOVA) and diagnostics

ANOVA was performed to analyze the fitting and significance of the model after model development [33]. Table 5 presents the
ANOVA results of the UCS model at the curing age of 7 days.
The F-value (F-value = 74.27) and p-value (p-value <0.0001) implied the significance of the UCS model. The highest p-value of all
the terms was 0.0002, indicating that the parameters were significant. Additionally, the “Lack of Fit” of 2.12 demonstrated that the
model’s lack of fit was non-significant compared to the pure error. The determination coefficient (R2) and adjusted R2 (Adj-R2) of the
mathematical model were close to 100%, indicating that the model was well fitted [34]. In the final statistical model, the high R2 (R2 =
0.9717) and Adj-R2 (Adj-R2 = 0.9586) with slight differences indicated the high consistency between experimental and predictive data
of the UCS. Moreover, the deviation between the Adj-R2 and Predicted R2 was under 0.2, showing reasonable agreement. The Adeq
Precision (signal to noise ratio) of 26.0035 was desirable for further explaining that this model can be used to navigate the design
space. The ratio of the standard deviation to the mean is defined as the coefficient of variation (C.V.%), which reflects the data
dispersion level. Herein, the value of C.V.% was found to be 4.32%, which denotes high experiment reliability due to the lower
dispersion from the mean. Consequently, the ANOVA results identified the significance of the final UCS statistical model, allowing for
performing diagnostics of the model.
Fig. 10 combines diagnostic plots, including the Box–Cox plot, the normal plot of residuals, Cook’s distance plot, leverage plot,
residuals vs. predicted response plot, and predicted vs. actual plot. The Box–Cox method provides information to obtain a more
suitable transformation. As shown in Fig. 10(a), low and high confidence interval values for lambda were − 0.35 and 1.98, respectively.
The confidence interval for lambda included 1, and thus, there was no recommendation to transform the data. The points were
distributed along a straight line in the normal plot of the residuals (Fig. 10(b)), indicating that the residuals were normally distributed.

Table 5
Analysis of variance (ANOVA) results of the UCS model cured for 7 days.

Source Sum of Squares Degrees of Freedom Mean Square F-value p-value Results

Model 5.92 6 0.9860 74.27 <0.0001 significant

Linear Mixture 4.36 3 1.45 109.59 <0.0001 significant
AD 0.7319 1 0.7319 55.14 <0.0001 significant
BC 0.3350 1 0.3350 25.23 0.0002 significant
CD 0.3714 1 0.3714 27.98 0.0001 significant
Residual 0.1726 13 0.0133
Lack of Fit 0.1333 8 0.0167 2.12 0.2119 insignificant
Pure Error 0.0393 5 0.0079
Cor Total 6.09 19
R2 = 0.9717 Adj-R2 = 0.9586 Pre-R2 = 0.9372 Adeq Precision = 26.0035 C.V.% = 4.32

8
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 10. Diagnostic plots including (a) Box–Cox plot, (b) normal plot of residuals, (c) Cook’s distance plot, (d) leverage plot, (e) residuals vs. predicted response plot,
and (f) predicted vs. actual plot.

Cook’s distance plot was obtained by evaluating the influence of the omission of one specific group from the model to facilitate the
detection of outliers. As shown in Fig. 10(c), the points were uniformly distributed, and there were no outliers due to the experimental
error. The uniformly distributed points in Fig. 10(d) and (e) were randomly scattered, supporting the result of the Cook’s distance plot.
Moreover, the outliers were ruled out in both Fig. 10(d) and (e). Fig. 10(f) shows the predicted vs. actual value of the UCS, which
revealed that the experimental UCS varied from 1.72 to 3.61 MPa whereas the predicted UCS varied from 1.76 to 3.54 MPa. The

9
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

analysis of the predicted vs. actual plot supported the high R2 value. Overall, the results of the ANOVA and diagnostics demonstrated
that the unconfined compressive strength model was representative of the experimental data.

3.3. Effects of mixture constituents on UCS of stabilized soils

The impact of the mixture constituents on the UCS of stabilized soil cured for 7 days was examined based on the Pearson correlation
coefficient (Fig. 11) and response surfaces (Fig. 12). Fig. 11 shows the Pearson correlation between the mixture components and UCS at
the curing age of 7 days. The Pearson correlation coefficient is characterized as the ratio of the covariance to the standard deviation,
where as the absolute value becomes to closer 1, it indicates a stronger correlation. The correlation values are displayed in the upper
triangular, and the corresponding colors are shown in the lower triangular. The correlation coefficients of 0.44 and 0.49 respectively
indicated that the BFS and PG were strongly positively correlated with the UCS. Within the constraints of the study, the UCS increased
as the mixed amounts of the BFS and PG increased.
Fig. 12(a) and (c) show that the UCS increased slowly with the increase in the content of component B and the decrease in the
content of component D under the constrained conditions of this experiment. The tendency of the UCS to increase with decreasing
content of component B and increasing content component A is shown in Fig. 12(c) and (d). Combined with Fig. 11, the correlation
coefficients of − 0.55 indicated that the SS was negatively correlated with the UCS. Thus, the steel slag was supposed to make little
contribution to the initial strength of the solidified soil. The minerals in steel slag including the silicate phase, RO phase (a continuous
solid solution of divalent metal oxide) and free calcium oxide, were speculated to lead a lower early activity [35].
The increase in the content of component A corresponded to the decrease in the content of component D, leading to a noticeable
increment of the UCS in Fig. 12(b) and (c). After that, the variation of the UCS tended to decrease. The result indicated that the blast
furnace slag could replace cement to some extent, and there was an optimal ratio of the two components.

3.4. Optimization and validation

The primary objective of this investigation was to optimize the composite material formula in the stabilized soil. The optimized
formula is shown in Table 6. In this study, the maximization of the UCS of stabilized soil was performed separately using the hill
climbing algorithm (HCA) and genetic algorithm (GA).
The HCA is a local searching mathematical optimization technique [36–38]. In this algorithm, the locally optimal solution is
obtained by searching the adjacent space of the current solution until reaching the local optimal solution [39,40]. Notably, the nu­
merical optimization algorithm in Design-Expert 12.0 was the hill climbing algorithm.
The genetic algorithm (GA) is a metaheuristic optimization approach to generate global optimal solutions by combining the
mechanism of genetics and evolution [41–45]. The GA was proposed by John Holland, inspired by Darwinian evolution [46,47]. The
flowchart of the algorithm is presented in Fig. 3, including the coding, reproduction process (selection, crossover, and mutation),
fitness calculation, and decoding [48–50]. The genes on the chromosomes are commonly encoded in a binary string in the GA [51]. The
GA searches for the optimal individuals under the process of selection, crossover, and mutation. After random inheritance and mu­
tation in the population P(t), the fittest form of a new generation P (t+1) is obtained. Then, the successive iterations of the new
generation retain the fittest genes. This process is interpreted as the genetic survival of the fittest [52].
In this study, the prediction model (Eq. (5)) was obtained after model training, variance analysis, and diagnostics. The constrained
boundary conditions of each variable in the DOD can be considered in the GA, but the quantitative relationship between each of the
variables cannot be taken into account simultaneously. Thus, a new function term considering the quantitative relationship between
the variables was introduced to the additional term of the fitness function to tackle the problem of the GA’s limitation in mixture
design. The fitness function was obtained by transforming the prediction model (Eq. (5)) into a negative function and adding a function
term expressing quantitative constraints. The function is shown as follows:
fun(xi ) = 0.089x1 − 0.158x2 − 0.466x3 + 0.275x4 − 0.008x1 x4 + 0.006x2 x3
(6)
+0.088x3 x4 + 1000 × abs(x1 + x2 + x3 + x4 − 99.9)

The quantitative relationship that the sum of the variables equals 100% was taken into account in the last term in Eq. (6). Conse­
quently, the minimum value of the fitness function demonstrated the best fitness in the study. After the establishment of the initial
population (initial population size = 1000) and the fitness function, the reproduction procedure was iterated until the optimal
chromosome was found.

Fig. 11. Pearson correlation between mixture components and UCS.

10
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 12. Two-dimensional contours and three-dimensional surface plots with fixed contents of SP (E = 0.1 wt%) (A: BFS, B: SS, C: PG, and D: P.O 42.5): (a) A = 53.82
wt%, (b) B = 10.36 wt%, (c) C = 10.36 wt, and (d) D = 25.36 wt%.

Table 6
Optimization of composite materials in stabilized soil.

Variables Units Lower Upper Target

BFS wt.% 40 60 in range

SS wt.% 0 15 in range
PG wt.% 0 15 in range
P.O 42.5 wt.% 15 30 in range
UCS MPa 1.72 3.61 maximize

11
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

The formula optimization of the composite materials in stabilized soils was carried out separately by the GA and HCA. The optimal
solution of composite materials formula in stabilized soil by the DOD and Hybrid DOD-GA along with validated results are shown in
Table 7. The hybrid DOD-GA approach predicted the maximum UCS of 3.46 MPa when the BFS was 56%, SS was 13.52%, PG was
14.87%, P.O 42.5 was 15.51%, and SP was 0.1%. In the contrast, the DOD predicted the maximum UCS of 3.45 MPa when the BFS was
59.75%, SS was 3.07%, PG was 14.81%, P.O 42.5 was 22.27%, and SP was 0.1%. The UCS of the optimal solution given by the DOD
technique was 3.12 MPa, with a mean absolute percentage error (MAPE) of 10.56% and a root mean square error (RMSE) of 0.34.
Furthermore, the optimal solution of the hybrid DOD-GA approach was verified, with a UCS value of 3.70 MPa, an MAPE of 6.99%, and
an RMSE of 0.24. Both the MAPE and RMSE of the experiment results of the hybrid DOD-GA technique were lower than those of the
DOD approach. Furthermore, the strength was greater than that obtained by the DOD technique. This demonstrated the better pre­
diction ability of the hybrid DOD-GA technique for the UCS.

4. Evaluation
The formula from the hybrid DOD-GA technique was 56% BFS, 13.52% SS, 14.87% PG, 15.51% P.O 42.5, and 0.1% SP. A group of
cement–soil samples was prepared to evaluate the properties of the new solidified soil. The proportion of the curing agent added to the
wet silt was 20%. The experimental macroscopic and microscopic characteristics of the two types of stabilized soil were examined
based on the UCS, SEM, and XRD results, as shown in Figs. 13–15 respectively.

4.1. UCS evaluation

Fig. 13 shows the UCS load curve of the optimal stabilized soil and cement soil after 7 days of aging. It is apparent that the optimal
stabilized soil and cement soil had significant peaks at 3.70 and 2.53 MPa, respectively. During the loading process, the stress grew
rapidly with the increase in the strain. Then, the stress–strain curve increased and decreased slowly near the peak values. After the
stress grew to the peak values, the specimen was damaged, and the curve went into a steep drop phase. In general, the optimal sta­
bilized soil exhibited a greater compressive strength and breaking strain than the cement soil.

4.2. Hydration characteristics (X-ray diffraction and scanning electron microscopy)

The crystalline phases of the stabilized soils were examined by XRD tests and analyzed with the Jade Software, and the results are
plotted in Fig. 14. The diffractogram revealed that the stabilized soils were composed of C-S-H, quartz, albite, orthoclase, calcite,
clinochlore, muscovite, cordierite, dolomite, and portlandite. In the study, the dry temperature of 110◦ C may cause decomposition of a
small amount of hydration products that decompose incompletely. Esteves [53] reported that the de-hydroxylation of C-S-H phased up
to 150 ◦ C, and the loss of weakly bind water on the gel solid took place at approximately 110 ◦ C. Furthermore, Liu. et al. [54] reported
the dehydration of C–S–H gel at around 120 ◦ C. The C-S-H was detected at the diffraction angles of 28.6◦ and 30.3◦ . The results were
consistent with the previous studies. In contrast to the results in Fig. 14(b), the portlandite is not evident in Fig. 14(a). Portlandite was
the product of the hydration reactions in the stabilized soils and decomposed between 400 ◦ C and 500 ◦ C [55,56]. The main hydration
reactions in the cement-stabilized soil are as follows:
3CaO · SiO2 + nH2 O→xCaO · SiO2 ·(n − 3 + x)H2 O + (3 − x)Ca(OH)2 (7)

2CaO · SiO2 + nH2 O → xCaO · SiO2 · (n − 2 + x)H2 O + (2 − x)Ca(OH)2 (8)

3CaO · Al2 O3 + nH2 O → 3CaO · Al2 O3 · nH2 O (9)

Portlandite was detected in the cement-stabilized soil. In addition to Eqs. (7)–(9), the hydration reactions of new curing-agent-
stabilized soil also included the following [57]:
/
CaSO4 · 1 2H2 O + 3/2H2 O → CaSO4 · 2H2 O (10)

3Ca(OH)2 + Al2 O3 +3CaSO4 · 2H2 O + 23H2 O → 3CaO · Al2 O3 · 3CaSO4 · 32H2 O (11)

It speculated that the reason for the absence of portlandite in the new curing agent-stabilized soil was that the hydration products of
P.O 42.5 and BFS reacted with PG to generate ettringite. In previous studies, the main mass loss of ettringite was concentrated in 90 ◦ C
and 450 ◦ C in the [58–60]. However, ettringite was not detected in the new curing agent stabilized soil possibly due to ettringite being
transformed into amorphous product after drying [61]. In addition, Bragg’s equation was introduced to calculate the interplanar
spacing of the C-S-H [62]. The interplanar spacing of the C-S-H in the cement silt was greater than that of the stabilized silt obtained by
the DOD-GA method, indicating a more compact structure in the optimal solidified silt.
The -COOH in the SP yielded strong adsorption, and the cement particles were dispersed due to electrostatic repulsion. The free
water trapped in the cement condensate was released. As shown in Eqs. (10) and (11), the free water was converted into bound water in
the formation of ettringite, leading to the strength increment of the new curing-agent-solidified soil. Based on Fig. 15, the clubbed
hydrate and acicular hydrate were speculated to be ettringite. Although the XRD could not detect ettringite being transformed into
amorphous product during drying, the initial form of ettringite in the cementitious matrix was retained [61]. The clubbed and acicular
ettringite embedded in the flocculent hydrate formed a spatial network structure. At 1000 × magnification, the particles of the new
curing-agent-stabilized silt were denser and more ordered than those of the cement-stabilized silt. Locally magnified to 3000 × , the
new curing-agent-stabilized silt developed a large amount of clubbed ettringite and flocculent hydrate (C-S-H) that filled the pores,

12
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Table 7
Optimal solution and validated results of composite material formula in stabilized soil by D-optimal design (DOD) and Hybrid D-optimal design–genetic algorithm
(DOD-GA).

Optimization methodology Variables Predicted response Validated response RMSE MAPE

A B C D E

DOD 59.75 3.07 14.81 22.27 0.1 3.45 3.12 0.34 10.56
Hybrid DOD-GA 56.00 13.52 14.87 15.51 0.1 3.46 3.70 0.24 6.99

Fig. 13. UCS load curves of two types of stabilized soil.

Fig. 14. Phase identification of the stabilized soils cured for 7 days: (a) optimal stabilized soil and (b) cement-stabilized soil.

which significantly improved the integrity (Fig. 15(a)). As for the cement-stabilized silt, there were small quantities of flocculent
hydrate and acicular crystals distributed on the surface, forming a weak reticular structure. The particles were observed to be
distributed in a disordered fashion, and the cement-stabilized silt particles exhibited weak cementation with more pores (Fig. 15(b)).
In summary, the regularity of the microstructure and macroscopic strength of the two kinds of solidified soil was fairly high
consistency. The new curing agent was more successful in enhancing the microstructure and strength of the silt than the cement was.

4.3. Life cycle assessment (LCA) analysis—global warming potential

According to the objectives of this study, the environmental impact of the optimal formulation curing agent from cradle to gate was
quantified, and the global warming potential (GWP) was selected as the environmental impact assessment indicator for life cycle
assessment. The functional unit of this study was 1 ton of curing agent. The environmental impact of the solid waste has been
considered in a large number of LCA studies, and solid waste is no longer regarded as industrial waste but as an industrial byproduct.
Therefore, the economic value allocation method was used to quantify the environmental impact of slag, steel slag, and phospho­
gypsum. The GWPs of the raw materials are depicted in Table 8. A truck with a 30-ton load was used for transportation, and its fuel
consumption for 100 km was 37.5 L [63]. The default value of the diesel CO2 was obtained from IPCC 2006 [64], and the calorific value

13
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Fig. 15. SEM images of the solidified soils at 1000 × and 3000 × magnification: (a) optimal stabilized soil and (b) cement-stabilized soil.

of diesel was obtained from the China Energy Statistics Manual [65]. The transportation distance was defined as the distance between
the raw material sources and the factory, and GWPs in the transportation stage are shown in Table 9.
The PG was transported to the plant and calcined at a temperature of 120◦ C-180◦ C [66]. The calcination reaction equation is as
follows:
120∘ C− 180∘ C
CaSO4 ⋅ 2H2 O ̅̅̅̅̅→ CaSO4 ⋅ 1 / 2H2 O + 3 / 2H2 O (12)
The theoretical enthalpies and GWPs of the PG calcination at different temperatures are shown in Table 10. The energy conversion
rate and the calcination conditions were set to 10% and 150◦ C for 1 h in this study, respectively [66]. The BFS, SS, and P.O 42.5 were
poured into the milling equipment for processing after the calcination of PG. With the 2100 mm (diameter) × 3000 mm (length) Red
Star ball mill equipment, the required power of the 1-ton curing agent was 7.29 kW•h. According to the survey statistics of the China
National Climate Strategy Center, the grid emission factor in east China is 0.7921 t CO2/MW•h. The GWPs of the manufacturing stage
are shown in Table 11.
Fig. 16 shows the environmental impact data of the new curing agent in each stage. It demonstrates that the GWP value of the raw
material production stage of the new curing agent was the largest proportion of the total LCA (i.e., 92.2%), followed by the
manufacturing stage (i.e., 4.2%). The P.O 42.5 and BFS made a significant contribution to the GWP of the raw material production
stage. As can be seen from Table 12, the GWP of the 1-ton new curing agent was 221.853 kg, while that of 1 ton of cement was 3.38
times that of the former at 750 kg.
Overall, the environmental impact of the new curing agent production was mainly from the raw material production stage, ac­
counting for 92.2%, which was much less than that of cement production. Furthermore, the strength of the new curing-agent-stabilized
soil was 3.70 MPa at the curing age of 7 days, compared to 2.53 MPa for cement, further demonstrating the superiority of the new
curing-agent-stabilized soil over cement soil in terms of the strength and environmental benefits.

Table 8
Global warming potentials (GWPs) of the production of the raw materials.

Raw materials 1 ton of optimal soil hardening agent/kg GWP/kg CO2-eq References

BFS 560 83.440 [67]

SS 135.2 1.091 [68]
PG 148.7 2.974 [69]
P.O 42.5 155.1 113.999 [70]
SP 1 3.090 [71]
Sum 1000 204.594

14
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Table 9
GWPs of the transportation stage.

Raw 1 ton of optimal soil Distance/ Fuel consumption of 100-km truck Default value of Calorific value of GWP/kg
materials hardening agent/kg km with 30-ton load CO2/kg•GJ− 1 China/kcal•L− 1 CO2-eq

BFS 560 300 37.5 74100 8800 5.733

SS 135.2 200 37.5 74100 8800 0.923
PG 148.7 200 37.5 74100 8800 1.015
P.O 42.5 155.1 50 37.5 74100 8800 0.265
SP 1 50 37.5 74100 8800 0.002
Sum 1000 / / / / 7.938

Table 10
Theoretical enthalpies and GWPs of PG calcination at different temperatures.

Raw Calcination Theory of enthalpy/ Energy exchange Calorific value of CO2 default value of GWP/kg
materials temperature/◦ C kJ•mol− 1 efficiency anthracite/kg•GJ− 1
anthracite/kcal•L− 1 CO2-eq

PG 120 4.960 0.1 7500 98300 3.393

130 5.034 0.1 7500 98300 3.444
140 5.107 0.1 7500 98300 3.494
150 5.177 0.1 7500 98300 3.542
160 5.246 0.1 7500 98300 3.589
170 5.313 0.1 7500 98300 3.635
180 5.379 0.1 7500 98300 3.680
190 5.443 0.1 7500 98300 3.724
200 5.507 0.1 7500 98300 3.768

Table 11
GWPs of the manufacturing stage.

Raw materials 1 ton of optimal soil hardening agent/kg GWP of calcination/kg CO2 eq Grinding and mixing/kg CO2 eq GWP/kg CO2 eq

BFS 560 / 0.896 0.896

SS 135.2 / 3.237 3.237
PG 148.7 3.542 0.781 4.323
P.O 42.5 155.1 / 0.859 0.859
SP 1 / 0.006 0.006
Sum 1000 3.542 5.779 9.321

Fig. 16. Environmental impact data of each stage of the new curing agent.

5. Conclusions
In this study, 20 groups of mixture designs were prepared by the D-optimal method to explore the effects of BFS, SS, PG, and P.O
42.5 on the solidified soil’s UCS. The following conclusions can be drawn:
(1) The BFS and PG exhibited a positive effect on the UCS within a certain range, and the SS made little contribution to the sta­
bilized soil’s UCS in the early stage. The maximum UCS was achieved at an optimal content ratio of the blast furnace slag to the
cement. The ANOVA results and diagnostics demonstrated that the Scheffé quadratic regression model was competitive in
predicting the UCS of the stabilized soil.

15
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

Table 12
GWPs of 1 ton of curing agent.

Type of curing agent Raw materials GWP/kg CO2-eq

Raw materials production stage Transportation stage Manufacturing stage Total

New curing agent BFS 83.440 5.733 0.896 90.069

SS 1.091 0.923 3.237 5.251
PG 2.974 1.015 4.323 8.312
P.O 42.5 113.999 0.265 0.859 115.123
SP 3.090 0.002 0.006 3.098
Sum 204.594 7.938 9.321 221.853
Cement P.O 42.5 / / / 750

(2) The verified strength of the solidified soil optimized by the hybrid DOD-GA technique was greater than that by the DOD
technique (UCS = 3.70 MPa for DOD-GA and UCS = 3.12 MPa for DOD), with a lower MAPE and RMSE (MAPE = 6.99% and
RMSE = 0.24 for the DOD-GA; MAPE = 10.56% and RMSE = 0.34 for the DOD).
(3) The optimal stabilized soil exhibited a greater compressive strength and breaking strain than the cemented soil. The soil
particles stabilized by the new curing agent were denser and filled more orderly, with a large amount of clubbed ettringite and
flocculent hydrate (C-S-H) in the pores, which significantly improved the integrity. The features of the microstructure and
macroscopic strengths of the two kinds of solidified soil showed fairly high consistency.
(4) The GWP of 1 ton of cement was 3.38 times that of 1 ton of the new curing agent. The new curing agent is more capable of
achieving improved strength, microstructure, and environmental benefits than cement is.
Overall, the observed industrial waste can be considered to be a substitute for cement to mitigate global warming. The optimized
curing agent has significant potential for industrial waste recycling, greenhouse buffering, and natural resource conservation.
This study examined the early strengths of solidified soils. However, the long-term strength is also essential to guarantee the en­
gineering safety. In the future, the impact of the curing age, dosage, organic content, and soil properties on the stabilized soil will be
investigated systematically to provide more scientific guidance for engineering.

CRediT author statement

Xin Chen: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Writing - Original Draft.
Feng Yu: Conceptualization, Resources, Supervision, Writing - Review & Editing, Funding acquisition.
Jing Yu: Validation, Investigation.
Shuaikang Li: Validation, Investigation.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

Data will be made available on request.

Acknowledgment
This work was supported by the National Natural Science Foundation of China (52178365) and the provincial research grant of
Zhejiang, China (2023C03142).

References
[1] IEA (International Energy Agency), Global energy review: CO2 emissions in 2021. https://www.iea.org/reports/global-energy-review-co2-emissions-in-2021-2,
2022. (Accessed 22 August 2022).
[2] IEA (International Energy Agency), Cement. https://www.iea.org/reports/cement, 2022. (Accessed 22 August 2022).
[3] P. Choeycharoen, W. Sornlar, A. Wannagon, A sustainable bottom ash-based alkali-activated materials and geopolymers synthesized by using activator solutions
from industrial wastes, J. Build. Eng. 54 (2022), 104659, https://doi.org/10.1016/j.jobe.2022.104659.
[4] I.S. Abbas, M.H. Abed, H. Canakci, Development and characterization of eco-and user-friendly grout production via mechanochemical activation of slag/rice
husk ash geopolymer, J. Build. Eng. 63 (2023), 105336, https://doi.org/10.1016/j.jobe.2022.105336.
[5] E. Gonzalez-Tolivia, S. Collado, P. Oulego, M. Diaz, BOF slag as a new alkalizing agent for the stabilization of sewage sludge, Waste Manage. (Tucson, Ariz.) 153
(2022) 335–346, https://doi.org/10.1016/j.wasman.2022.09.009.
[6] J. Jin, Z. Qin, H. Yang, S. Zuo, C. Song, Designing a composite solidification agent to improve mechanical properties and sulfate resistance of construction clay-
residue, J. Build. Eng. 56 (2022), 104753, https://doi.org/10.1016/j.jobe.2022.104753.
[7] H. Miraki, N. Shariatmadari, P. Ghadir, S. Jahandari, Z. Tao, R. Siddique, Clayey soil stabilization using alkali-activated volcanic ash and slag, J. Rock Mech.
Geotech. Eng. 14 (2) (2022) 576–591, https://doi.org/10.1016/j.jrmge.2021.08.012.
[8] L. Liu, A.N. Zhou, Y.F. Deng, Y.J. Cui, Z. Yu, C. Yu, Strength performance of cement/slag-based stabilized soft clays, Construct. Build. Mater. 211 (2019)
909–918, https://doi.org/10.1016/j.conbuildmat.2019.03.256.

16
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

[9] N.J. Jiang, Y.J. Du, K. Liu, Durability of lightweight alkali-activated ground granulated blast furnace slag (GGBS) stabilized clayey soils subjected to sulfate
attack, Appl. Clay Sci. 161 (2018) 70–75, https://doi.org/10.1016/j.clay.2018.04.014.
[10] J.S. Li, L. Chen, B.J. Zhan, L. Wang, C.S. Poon, D.C.W. Tsang, Sustainable stabilization/solidification of arsenic-containing soil by blast slag and cement blends,
Chemosphere 271 (2021), 129868, https://doi.org/10.1016/j.chemosphere.2021.129868.
[11] M. Salimi, A. Ghorbani, Mechanical and compressibility characteristics of a soft clay stabilized by slag-based mixtures and geopolymers, Appl. Clay Sci. 184
(2020), 105390, https://doi.org/10.1016/j.clay.2019.105390.
[12] H.M. Ren, W.B. Liu, D.W. Zhang, Application of phosphogypsum to solidification of silty soil: mechanical properties and microstructure, Mech. Adv. Mater.
Struct. 29 (27) (2022) 6026–6038, https://doi.org/10.1080/15376494.2021.1971804.
[13] H. Al-Dakheeli, R. Bulut, G. Scott Garland, C.R. Clarke, Utilization of blast-furnace slag as a standalone stabilizer for high sulfate-bearing soils, J. Mater. Civ.
Eng. 33 (10) (2021), 04021295, https://doi.org/10.1061/(ASCE)MT.1943-5533.0003880.
[14] S. Meskini, T. Remmal, H. Ejjaouani, A. Samdi, Formulation and optimization of a phosphogypsum-fly ash-lime composite for road construction: a statistical
mixture design approach, Construct. Build. Mater. 315 (2022), 125786, https://doi.org/10.1016/j.conbuildmat.2021.125786.
[15] H. Scheffé, Experiments with mixtures, J R Stat Soc Series B Stat Methodol 20 (2) (1958) 344–360, https://doi.org/10.1111/j.2517-6161.1958.tb00299.x.
[16] X.P. Wang, R. Yu, Q.L. Song, Z.H. Shui, Z. Liu, S. Wu, D.S. Hou, Optimized design of ultra-high performance concrete (UHPC) with a high wet packing density,
Cement Concr. Res. 126 (2019), 105921, https://doi.org/10.1016/j.cemconres.2019.105921.
[17] C. Varanda, I. Portugal, J. Ribeiro, A.M.S. Silva, C.M. Silva, Optimization of bitumen formulations using mixture design of experiments (MDOE), Construct.
Build. Mater. 156 (2017) 611–620, https://doi.org/10.1016/j.conbuildmat.2017.08.146.
[18] Y. Sun, R. Yu, S.Y. Wang, Y.X. Zhou, M. Zeng, F.J. Hu, Z.H. Shui, B.Y. Rao, S. Yuan, Z.L. Luo, S. Ma, Development of a novel eco-efficient LC2 conceptual cement
based ultra-high performance concrete (UHPC) incorporating limestone powder and calcined clay tailings: design and performances, J. Clean. Prod. 315 (2021),
128236, https://doi.org/10.1016/j.jclepro.2021.128236.
[19] A.K. Xu, R. Li, H.M. Chang, Y.J. Xu, X. Li, G. Lin, Y. Zhao, Artificial neural network (ANN) modeling for the prediction of odor emission rates from landfill
working surface, Waste Manage. (Tucson, Ariz.) 138 (2022) 158–171, https://doi.org/10.1016/j.wasman.2021.11.045.
[20] D.W. Jiao, C.J. Shi, Q. Yuan, X.P. An, Y. Liu, Mixture design of concrete using simplex centroid design method, Cem. Concr. Compos. 89 (2018) 76–88, https://
doi.org/10.1016/j.cemconcomp.2018.03.001.
[21] F. Zahiri, H. Eskandari-Naddaf, Optimizing the compressive strength of concrete containing micro-silica, nano-silica, and polypropylene fibers using extreme
vertices mixture design, Front. Struct. Civ. Eng. 13 (4) (2019) 821–830, https://doi.org/10.1007/s11709-019-0518-6.
[22] X.D. Wang, F. Tsung, W.D. Li, D.D. Xiang, C. Cheng, Optimal space-filling design for symmetrical global sensitivity analysis of complex black-box models, Appl.
Math. Model. 100 (2021) 303–319, https://doi.org/10.1016/j.apm.2021.08.015.
[23] E. Ghafari, H. Costa, E. Julio, Statistical mixture design approach for eco-efficient UHPC, Cem. Concr. Compos. 55 (2015) 17–25, https://doi.org/10.1016/j.
cemconcomp.2014.07.016.
[24] M. Venkatesan, Q. Zaib, I.H. Shah, H.S. Park, Optimum utilization of waste foundry sand and fly ash for geopolymer concrete synthesis using D-optimal mixture
design of experiments, Resour. Conserv. Recycl. 148 (2019) 114–123, https://doi.org/10.1016/j.resconrec.2019.05.008.
[25] Ministry of Housing and Urban-Rural Development of the People’s Republic of China, JGJ/T 233—2011. Specification for Mix Proportion Design of Cement Soil,
2011 (in Chinese).
[26] Ministry of Housing and Urban-Rural Development of the People’s Republic of China, JGJ/T 303-2013. Technical Specification for Trench Cutting Re-mixing
Deep Wall, 2013 (in Chinese).
[27] H. Huang, T. Wang, B. Kolosz, J. Andresen, S. Garcia, M.X. Fang, M.M. Maroto-Valer, Life-cycle assessment of emerging CO2 mineral carbonation-cured concrete
blocks: comparative analysis of CO2 reduction potential and optimization of environmental impacts, J. Clean. Prod. 241 (2019), 118359, https://doi.org/
10.1016/j.jclepro.2019.118359.
[28] L.F. Li, T.C. Ling, S.Y. Pan, Environmental benefit assessment of steel slag utilization and carbonation: a systematic review, Sci. Total Environ. 806 (2022),
150280, https://doi.org/10.1016/j.scitotenv.2021.150280.
[29] T. Wang, Z. Yi, J. Song, C. Zhao, R. Guo, X. Gao, An industrial demonstration study on CO2 mineralization curing for concrete, iScience 25 (5) (2022), 104261,
https://doi.org/10.1016/j.isci.2022.104261.
[30] B. Marmiroli, M. Venditti, G. Dotelli, E. Spessa, The transport of goods in the urban environment: a comparative life cycle assessment of electric, compressed
natural gas and diesel light-duty vehicles, Appl. Energy 260 (2020), 114236, https://doi.org/10.1016/j.apenergy.2019.114236.
[31] M.J. Anderson, P.J. Whitcomb, in: DOE Simplified: Practical Tools for Effective Experimentation, third ed., Productivity Press, New York, 2015.
[32] N.S. Pillai, P.S. Kannan, S.C. Vettivel, S. Suresh, Optimization of transesterification of biodiesel using green catalyst derived from Albizia Lebbeck Pods by
mixture design, Renew. Energy 104 (2017) 185–196, https://doi.org/10.1016/j.renene.2016.12.035.
[33] J.W. Baek, A.E.S. Choi, H.S. Park, Solidification/stabilization of ASR fly ash using Thiomer material: optimization of compressive strength and heavy metals
leaching, Waste Manage. (Tucson, Ariz.) 70 (2017) 139–148, https://doi.org/10.1016/j.wasman.2017.09.010.
[34] D.Q. Fan, R. Yu, Z.H. Shui, K.N. Liu, Y. Feng, S.Y. Wang, K.K. Li, J.H. Tan, Y.J. He, A new development of eco-friendly ultra-high performance concrete (UHPC):
towards efficient steel slag application and multi-objective optimization, Construct. Build. Mater. 306 (2021), 124913, https://doi.org/10.1016/j.
conbuildmat.2021.124913.
[35] X.L. Guo, H.S. Shi, Utilization of steel slag powder as a combined admixture with ground granulated blast-furnace slag in cement based materials, J. Mater. Civ.
Eng. 25 (12) (2013) 1990–1993, https://doi.org/10.1061/(ASCE)MT.1943-5533.0000760.
[36] I. Sohn, A robust complex network generation method based on neural networks, Physica A 523 (2019) 593–601, https://doi.org/10.1016/j.
physa.2019.02.046.
[37] G. Ryma, K. Mohamed-Khireddine, Genetic algorithm with hill climbing for correspondences discovery in ontology mapping, J. Inf. Technol. Res. 12 (4) (2019)
153–170, https://doi.org/10.4018/JITR.2019100108.
[38] A. Arram, M. Ayob, A. Sulaiman, Hybrid bird mating optimizer with single-based algorithms for combinatorial optimization problems, IEEE Access 9 (2021)
115972–115989, https://doi.org/10.1109/ACCESS.2021.3102154.
[39] P.P. Selvam, R. Narayanan, Random restart local search optimization technique for sustainable energy-generating induction machine, Comput. Electr. Eng. 73
(2019) 268–278, https://doi.org/10.1016/j.compeleceng.2018.11.023.
[40] A. Khajeh, Z. Hamzavi-Zarghani, A. Yahaghi, A. Farmani, Tunable broadband polarization converters based on coded graphene metasurfaces, Sci. Rep. 11 (1)
(2021) 1296, https://doi.org/10.1038/s41598-020-80493-w.
[41] L. Yue, H.W. Li, Y.N. Li, C.Y. Jin, Optimum design of high-strength concrete mix proportion for crack resistance using artificial neural networks and genetic
algorithm, Front. Mater. 7 (2020), 590661, https://doi.org/10.3389/fmats.2020.590661.
[42] A.A.S. Javid, H. Naseri, M.A.E. Ghasbeh, Estimating the optimal mixture design of concrete pavements using a numerical method and meta-heuristic algorithms,
JST-T MECH ENG. 45 (2) (2021) 913–927, https://doi.org/10.1007/s40996-020-00352-6.
[43] C. Su, J.W. Bai, Structural optimization of ship lock heads during construction period considering concrete creep, Math. Probl Eng. 2020 (2020), 5495202,
https://doi.org/10.1155/2020/5495202.
[44] R.A. Mozumder, S.M. Laskar, A.I. Laskar, Optimization of geometrical features of a vane concrete rheometer using genetic algorithm, Arabian J. Sci. Eng. 46
(11) (2021) 11279–11290, https://doi.org/10.1007/s13369-021-05781-7.
[45] J.S. Tu, Y.Z. Liu, M. Zhou, R.X. Li, Prediction and analysis of compressive strength of recycled aggregate thermal insulation concrete based on GA-BP
optimization network, J. Eng. Des. Technol. 19 (2) (2021) 412–422, https://doi.org/10.1108/JEDT-01-2020-0022.
[46] L.D. Machado, A.C. Fernandes, Moonpool dimensions and position optimization with Genetic Algorithm of a drillship in random seas, Ocean Eng. 247 (2022),
110561, https://doi.org/10.1016/j.oceaneng.2022.110561.
[47] A. Kandiri, F. Sartipi, M. Kioumarsi, Predicting compressive strength of concrete containing recycled aggregate using modified ANN with different optimization
algorithms, Appl. Sci. 11 (2) (2021) 485, https://doi.org/10.3390/app11020485.

17
X. Chen et al. Journal of Building Engineering 72 (2023) 106611

[48] M. Zafar, N.V. Vinh, S.K. Behera, H.S. Park, Ethanol mediated As(III) adsorption onto Zn-loaded pinecone biochar: experimental investigation, modeling, and
optimization using hybrid artificial neural network-genetic algorithm approach, J. Environ. Sci. (China) 54 (2017) 114–125, https://doi.org/10.1016/j.
jes.2016.06.008.
[49] X.Y. Wang, H.S. Lee, Effect of global warming on the proportional design of low CO2 slag-blended concrete, Construct. Build. Mater. 225 (2019) 1140–1151,
https://doi.org/10.1016/j.conbuildmat.2019.07.134.
[50] X.Y. Wang, Simulation for optimal mixture design of low-CO2 high-volume fly ash concrete considering climate change and CO2 uptake, Cem. Concr. Compos.
104 (2019), 103408, https://doi.org/10.1016/j.cemconcomp.2019.103408.
[51] A. Duary, M.S. Rahman, A.A. Shaikh, S.T.A. Niaki, A.K. Bhunia, A new hybrid algorithm to solve bound-constrained nonlinear optimization problems, Neural
Comput. Appl. 32 (16) (2020) 12427–12452, https://doi.org/10.1007/s00521-019-04696-7.
[52] Z. Kaseb, M. Rahbar, Towards CFD-based optimization of urban wind conditions: comparison of genetic algorithm, particle swarm optimization, and a hybrid
algorithm, Sustain. Cities Soc. 77 (2022), 103565, https://doi.org/10.1016/j.scs.2021.103565.
[53] L.P. Esteves, On the hydration of water-entrained cement–silica systems: combined SEM, XRD and thermal analysis in cement pastes, Thermochim. Acta 518
(1–2) (2011) 27–35, https://doi.org/10.1016/j.tca.2011.02.003.
[54] M. Liu, J. Lei, L. Guo, X. Du, J. Li, The application of thermal analysis, XRD and SEM to study the hydration behavior of tricalcium silicate in the presence of a
polycarboxylate superplasticizer, Thermochim. Acta 613 (2015) 54–60, https://doi.org/10.1016/j.tca.2015.05.020.
[55] N.J. Jiang, Y.J. Du, S.Y. Liu, M.L. Wei, S. Horpibulsuk, A. Arulrajah, Multi-scale laboratory evaluation of the physical, mechanical, and microstructural
properties of soft highway subgrade soil stabilized with calcium carbide residue, Can. Geotech. J. 53 (3) (2015) 373–383, https://doi.org/10.1139/cgj-2015-
0245.
[56] J. Wu, Z. Deng, Y. Deng, A. Zhou, Y. Zhang, Interaction between cement clinker constituents and clay minerals and their influence on the strength of cement-
based stabilized soft clay, Can. Geotech. J. 59 (6) (2022) 889–900, https://doi.org/10.1139/cgj-2021-0194.
[57] J. He, X.K. Shi, Z.X. Li, L. Zhang, X.Y. Feng, L.R. Zhou, Strength properties of dredged soil at high water content treated with soda residue, carbide slag, and
ground granulated blast furnace slag, Construct. Build. Mater. 242 (2022), 118126, https://doi.org/10.1016/j.conbuildmat.2020.118126.
[58] D. Yang, W. Sun, Z. Liu, Thermal decomposition kinetics of ettringite crystal, J. Chin. Ceram. Soc. 224 (12) (2007) 1641–1645+1656, https://doi.org/
10.14062/j.issn.0454-5648.2007.12.001 (in Chinese).
[59] Z.H. Ou, B.G. Ma, S.W. Jian, Comparison of FT-IR, thermal analysis and XRD for determination of products of cement hydration, Adv. Mater. Res. 168 (2011)
518–522. https://doi.org/10.4028/www.scientific.net/AMR.168-170.518.
[60] E. Qoku, T.A. Bier, T. Westphal, Phase assemblage in ettringite-forming cement pastes: a X-ray diffraction and thermal analysis characterization, J. Build. Eng.
12 (2017) 37–50, https://doi.org/10.1016/j.jobe.2017.05.005.
[61] S. Afflerbach, C. Pritzel, P. Hartwich, M.S. Killian, W. Krumm, Effects of thermal treatment on the mechanical properties, microstructure and phase composition
of an Ettringite rich cement, CEMENT 11 (2023), 100058, https://doi.org/10.1016/j.cement.2023.100058.
[62] X. Chen, F. Yu, Z. Hong, L. Pan, X. Liu, Y. Li, Comparative investigation on the curing behavior of GS-stabilized and cemented soils at macromechanical and
microstructural scales, J. Test. Eval. 51 (1) (2021), https://doi.org/10.1520/JTE20200631.
[63] Z.F. Yang, Research on carbon emission regulations of heavy commercial vehicles at home and abroad, Automobile Applied Technology 45 (16) (2020)
256–258, https://doi.org/10.16638/j.cnki.1671-7988.2020.16.083 (in Chinese).
[64] IPCC (Intergovernmental Panel on Climate Change), 2006 IPCC guidelines for national greenhouse gas inventories. https://www.ipcc-nggip.iges.or.jp/public/
2006gl/chinese/vol2.html, 2006. (Accessed 29 September 2022).
[65] National Bureau of, Statistics of the People’s Republic of China, Energy Statistics Workbook, China statistics Press, Beijing, 2010 (In Chinese).
[66] R.H. Geraldo, A.R.D. Costa, J. Kanai, J.S. Silva, J.D. Souza, H.M.C. Andrade, J.P. Goncalves, P.S.P. Fontanini, G. Camarini, Calcination parameters on
phosphogypsum waste recycling, Construct. Build. Mater. 256 (2020), 119406, https://doi.org/10.1016/j.conbuildmat.2020.119406.
[67] C. Chen, G. Habert, Y. Bouzidi, A. Jullien, A. Ventura, LCA allocation procedure used as an incitative method for waste recycling: an application to mineral
additions in concrete, Resour. Conserv. Recycl. 54 (12) (2010) 1231–1240, https://doi.org/10.1016/j.resconrec.2010.04.001.
[68] H.W. Kua, Integrated policies to promote sustainable use of steel slag for construction—a consequential life cycle embodied energy and greenhouse gas emission
perspective, Energy Build. 101 (2015) 133–143, https://doi.org/10.1016/j.enbuild.2015.04.036.
[69] Y. Zhou, C.Q. Xiao, S. Yang, H.Q. Yin, Z.Y. Yang, R. Chi, Life cycle assessment of feed grade mono-dicalcium phosphate production in China, a case study,
J. Clean. Prod. 290 (2021), 125182, https://doi.org/10.1016/j.jclepro.2020.125182.
[70] C. Li, S. Cui, Z. Nie, X. Gong, Z. Wang, N. Itsubo, The LCA of portland cement production in China, Int. J. Life Cycle Assess. 20 (1) (2015) 117–127, https://doi.
org/10.1007/s11367-014-0804-4.
[71] Y. Liang, X. Shi, C. Zhang, T. Zhang, X. Wang, Comprehensive evaluation of the performance and environmental Impact of fly ash geopolymer concrete,
Materials Reports 37 (2) (2023), 21060162 (in Chinese), http://kns.cnki.net/kcms/detail/50.1078.TB.20220221.0855.002.html.

18