Mechanical Properties and Drying Shrinkage Investigation of Alkali-Activated Mortar Using Waste Glass Powder

Alkali-activated mortar (AAM) is one of the products of waste glass recycling that exhibits promising potential for wide engineering applications such as the construction industry. In this study, recycled waste glass powder-based additives, namely, Silica Fume (SF) and Nano-SiO2 (NS), were investigated for their potential to enhance the mechanical properties (strength) and dryingshrinkage resistance of AAM. )e results indicated that 5.0% and 1.5% were the optimum SF and NS dosages, respectively, for optimizing AAM performance in terms of the compressive strength, flexural strength, and drying-shrinkage resistance. A prediction model, based on backpropagation (BP) neural network analysis, was also satisfactorily formulated and preliminarily validated for predicting the drying shrinkage of AAM containing SF or NS.


Introduction
Due to the worldwide increase in the use of glass products, the pursuit for waste glass recycling as a means of used glass disposal has similarly become an increasing trend [1][2][3]. e traditional landfilling of waste glass undesirably leads to environmental pollution and is unsustainable [4]. Different from the landfilling method, waste glass can be beneficially recycled to produce alkali-activated mortar (AAM) [5,6] or alkali-activated concrete (AAC) [7,8].
AAM and AAC often have better mechanical properties in the presence of waste glass powder (WGP) [9][10][11]. e literature indicates that the compressive strength of AAM [12] or AAC [13] could be improved significantly by blending with WGP. Due to the increased brittleness, however, the flexural strength growth of AAM or AAC has been reported to be insignificant when compared to the compressive strength growth rate. is increase in brittleness may exacerbate the risk of cracking, which ultimately poses a challenge and inhibits their (AAM and AAC) practical application and usage in the industry. e dry shrinkage of AAM is usually greater than that of ordinary Portland concrete (OPC). erefore, reducing shrinkage and cracking remains a key challenge [14]. Due to the potential reactivity of SiO 2 in WGP, it is effective in decreasing the early and long-term drying shrinkage of AAM [15]. When the alkali content was 8.31%, the addition of 14.57% WGP resulted in about 15.8% and 20.3% drying shrinkage reduction at 1 day and 120 days, respectively [16]. However, the drying shrinkage of AAM modified with WGP was still much higher than that of OPC [17].
In this study, the effects of Silica Fume (SF) and Nano-SiO 2 (NS) on the strength and dry shrinkage of AAM were comparatively evaluated. e correlations between compressive strength and curing age were analyzed. A prediction model for drying shrinkage analysis was established using the Backpropagation (BP) neural network [18].

Cementitious Material.
A composite cementitious material composed of slag powder (SP), SF, NS, and WGP (300 mesh) was used. Table 1 shows the chemical composition of each component in the composite cementitious material.

Sand.
Quartz sand (40∼70 mesh size) with a density of 3.65 g/cm 3 was used as the fine aggregate for all the mixdesigns.

Alkali Activator.
e alkaline activators used in this study comprised of reagent sodium hydroxide (NaOH) flakes (96% purity) and sodium silicate (the SiO 2 /Na 2 O ratio modulus was 2.3). NaOH and sodium silicate were mixed to obtain an alkali activator with a modulus of 1.3 (ratio of SiO 2 to Na 2 O).

Mix-Design Proportions.
e activator solution was cementitious material ratio of 0.50, and the sand to cementitious material ratio of 1 was kept constant for all the mix-designs. e Na 2 O concentration was kept at 6.0% (by mass of cementitious material in the alkaline activator solution). For AAM, SP was partially replaced by SF (5%, 10%, 15%, 20%, and 25%) and NS (0.5%, 1%, 1.5%, 2%, and 3%) by weight of the cementitious material. In the control group, the WGP and SP were kept at 30% and 70% by mass of the whole cementitious material, respectively. Table 2 presents the detailed AAM mix proportion.

Test Methods.
e compressive and flexural strength tests were conducted in accordance with the Chinese Test Method for Strength of Cement Mortar (GB/T17671-1999, IOS) [19]. For each mix, 40 mm × 40 mm × 160 mm prismatic specimens were prepared and tested. e specimens were cured under standard curing conditions (20 ± 2°C, relative humidity ≥ 95%) for 3, 7, 28, and 56 days, respectively. e dry shrinkage test of AAM was executed following the Chinese Hydraulic Concrete Test Standard SL352-2006 [20]. Prismatic specimens of 40 mm × 40 mm × 160 mm in dimensions were used. e specimens were demoulded after curing in the standard curing box for 48 hours. e specimens were thereafter placed into the dry shrinkage curing box (20 ± 2°C and 60 ± 5% relative humidity) after measuring the initial lengths. Measurements were conducted at ages of 3, 7, 28, and 56 days, respectively.
For all the tests, i.e., compressive strength, flexural strength, and shrinkage, three specimen replicates were used per test condition per mix-design proportion.

Test Results and Analysis
e test results are presented and analyzed in this section. e analysis includes the strength drying-shrinkage response behavior of SF and NS additives. Figure 1 shows the effects of SF dosage on the compressive and flexural strength of AAM. e results show that SF had an adverse effect on compressive strength. And the degree of this effect was more pronounced with an increase in the SF dosage.

Effects of SF and NS on Compressive and Flexural Strength.
As shown in Figure 1(a), when the SF content was 5% (S1) by weight of the cementitious material, the 28-day compressive strength increased by 2.6%. And when the SF content was 10%, 15%, 20%, and 25%, respectively, the 28day compressive strength decreased by 10.2%, 19.8%, 23%, and 26.15%, respectively. It indicated that the compressive strength decreased more rapidly once the SF content was higher than 10%. After 56 days of curing, the compressive strength containing 5% SF decreased by about 2.34% compared to the control group. When SF was 10%, 15%, 20%, and 25%, respectively, the compressive strength decreased by 15.74%, 21.3%, 22.05%, and 24.62%, respectively.
As shown in Figure 1(b), the 28-day flexural strength increased by 1.5% when SF was 5% (S1) by weight of the cementitious material. When the dosage reached 10%, 15%, 20%, and 25%, respectively, reductions of 12.3%, 13.8%, 13.8%, and 24.6% in the 28-day flexural strength were observed, respectively. At the age of 56 days, the flexural strength with 5% SF was increased by 1.5%. When the SF content was 10%, 15%, 20%, and 25%, the flexural strength was decreased by 15.5%, 16.9%, 18.3%, and 23.9%, respectively. However, the reduction in flexural strength was much lower than that of the compressive strength, indicating an obvious enhancement of the AAM toughness. But when the SF content was higher than 10%, the flexural strength decreased rapidly.
Overall, Figure 1 suggests that 5% (S1) is the optimum SF dosage for simultaneously maximizing both the AAM compressive and flexural strengths.
Because the amorphous SiO 2 in SF promotes the formation of gel [21], when SF content was 5%, it influenced the mechanical properties of AAM positively. When SF content was greater than 10%, it would cause a decrease in the calcium-silicon ratio in AAM. An excessively high content of active amorphous SiO 2 not only slowed the hydrolysis of Si-O-Si [22], but also increased the degree of silicate ion polymerization. Silicates tend to form monomers rather than polymers, thus reducing the polymerization degree of AAM.
At the same time, SF participated in the reaction, which caused water consumption, resulting in slow dissolution of the calcium-rich phase in SP, which is not conducive for the formation of stable polymerization products [23]. Consequently, when the SF content was greater than 10%, it adversely affected the mechanical performance of AAM and, hence, caused a decay in the strength magnitudes. Figure 2 shows the effect of NS content on the compressive and flexural strengths of AAM. e results show that NS could be contributing to both the compressive and flexural strength.
As shown in Figure 2(a), the compressive strength initially increased and then decreased as the NS dosage was increased. When 2% NS (NS4) by weight of cementitious material was added, the compressive strength at the age of 28 days and 56 days peaked at 82.07 MPa (with a 15.7% rise) and 89.05 MPa (with an 8.8% rise), respectively.
Similarly, the AAM flexural strength variation as a function of NS content was almost the same as that of the compressive strength trend. In the presence of 2% NS by weight of cementitious material, the 28-day and 56-day flexural strengths increased by 15.8% and 18.7%, respectively. e results indicated that the flexural strength was more efficiently enhanced by blending with 2% NS by weight of cementitious material.
Overall, Figure 2 suggests that 2% (NS4) is the optimum NS dosage for simultaneously maximizing both the AAM compressive and flexural strengths.
Based on the graphical trends in Figure 2, it is apparent that the mechanical properties of AAM had a clear relationship with the degree of mix polycondensation [24]. Due to its surface and nucleation effects, NS will promote the reaction and dissolution of the calcium-rich phase in the slag to form a stable polymerization product. With an increase in the NS content, the mechanical strength of AAM would also increase.
When the amount of NS is greater than 2% by weight of cementitious material, the NS size and surface effects are particularly stronger due to the electrostatic effects of NS and the highly active chemical bonds [25].
is leads to the phenomenon of agglomeration of NS, which cannot fully exert the surface of NS. e effect also makes part of the NS unreactive, resulting in a decrease in the mechanical strength of the AAM [26].

Effects of SF and NS on of the Bending-Compressive
Strength Ratio. e ratio of bending-compressive strength (Rt) is a vital index of the toughness of cementitious materials. e Rt values are illustrated in Figure 3 and indicate a gradual decline with age.
As shown in Figure 3(a), Rt with an increase in the SF dosage, indicating a decay in AAM toughness. When the SF content was 5% and 25% by weight of cementitious material, respectively, the Rt at 56 days was 8.7% and 6.3%, respectively.
As shown in Figure 3(b), for NS-reinforced AAM, the reduction rate of early Rt was faster than that of AAM in the absence of NS. With aging, the development of compressive and flexural strengths tended to be slow, but the flexural strength growth was higher than the compressive strength, thus increasing the overall toughness of AAM. When the NS dosage was increased from 0.5% to 3% by weight of cementitious material, respectively, the Rt at 56 days increased from 8.8% to 9.8%, respectively, indicating that NS could improve the AAM toughness.   Advances in Civil Engineering
SF can improve the reaction rate [21], which increases the water consumption and, ultimately, leads to a relatively high early drying shrinkage as evident in Figure 4 for 3 and 7 days curing ages.
When the SF content was less than 10%, the drying shrinkage was gradually lower than that of the control group due to the improvement in the matrix pore structure.
When the SF content was 10%, more low-density C-S-H gels were formed [27], resulting in higher drying shrinkage; thus, a maximum drying shrinkage at 56 days was observed. However, when the SF content was more than 10%, some of the SF particles did not participate in the hydration reaction, thereby filling in the AAM pores and, ultimately, reducing the drying shrinkage. e effects of NS on the drying shrinkage of AAM are shown in Figure 5. With the increase in the amount of NS, the 7-day drying shrinkage relative to the control group increased by −22.8%, −5%, 33.5%, 35.1%, and 69.2%, respectively. It can be seen from the 28-day and 56-day drying shrinkage values that the growth rate of the drying shrinkage values in the later period began to become placid. e results showed that the drying shrinkage of AAM increased with an increase in the NS content of about 0.5%∼3%. e surface and nucleation effects of NS help to promote the reaction rate of AAM [28]. As the amount of NS increases, the reaction makes the water consumption to increase, which leads to an increase in the early drying shrinkage. In this study, the amount of slag is the largest and the pore size in the matrix is mostly less than 10 nm or greater than 200 nm [29]. is is because the capillary pores are composed of mesopores and macropores, which is one of the reasons for the moisture migration pores between the matrix particles. e shrinkage value mainly depends on the water loss of the mesopores and pore size of the macropores [30]. With an increase in the degree of matrix polymerization and loss of water dispersion, the reaction slows down and the pore structure of NS becomes refined. So when the amount of NS particles is less than 1.5%, the postdrying shrinkage value is lower than that of the control group. When the amount of NS particles is greater than 1.5%, due to the agglomeration phenomenon [31], the NS cannot fully react, which may cause the degree of polymerization within the matrix to be different-resulting in voids and a drying shrinkage value greater than the control group.

Compressive Strength Prediction Based on Linear
Regression Analysis. It is a known fact that curing age plays a crucial role in the development of the compressive strength of AAM. However, the evolution of the compressive strength of the AAM is quite different from that of OPC. e relationships between age and compressive strength are illustrated in Figure 6.
As shown in Figure 6, R 2 (the coefficient of determination) is the fitting degree and defines the strength of the relationship between the variables, i.e., the higher the R 2 value, the stronger the relationship and the higher the prediction accuracy [32]. e fitting degree indexes in Figure 6 are all greater than 0.94 (i.e., R 2 ≥ 94%), which proves that the fitting degree is high and the correlation is        erefore, a prediction curve of compressive strength can be obtained for AAM containing SF or NS as follows: (1) e regression coefficients, a and b, are listed in Table 3. To verify the applicability of the prediction model, the calculated compressive strength was compared with the measured test data. As shown in Figure 6(i), the relative error of the compressive strength between the calculated value and the experimental data did not exceed 5%, which means that the model predictions were in agreement with the experimental data. e compressive strength model of AAM was obtained with relatively high reliability, i.e., R 2 > 90%. However, because the compressive strength is also affected by other factors such as the water-to-cementitious material ratio, curing conditions, and specimen size, further analysis was necessary to validate the model predictions.

Drying Shrinkage Prediction Model Based on BP Neural
Network. From the perspective of the drying shrinkage mechanism of alkali-activated materials, the pore structure theory [33,34] assumes that drying shrinkage was mainly caused by changes in its pore distribution and capillary tension, while the C-S-H gel theory [35] considers that the drying shrinkage of AAM was predominantly due to the low-density (LD) C-S-H gel, which causes drying shrinkage. erefore, it was considered that the factors affecting the drying shrinkage include the types of gelling materials, water-to-cementitious material ratio, curing conditions, and test block size [36]. At the same time, the factors influencing the drying shrinkage of AAM also had similar impacts on the compressive strength, suggesting the existence of some correlations. erefore, it was deemed possible to establish a prediction model for age, compressive strength, and drying shrinkage, respectively.
In order to effectively predict the size change caused by shrinkage, a nonparametric method was proposed. at is, to use BP neural network [18] to train the measured data and effectively establish a model to predict shrinkage. e drying shrinkage prediction model uses a multilayer Backpropagation algorithm, which relies on learning from the measured data and appropriately adjusting the weights of the neural network. BP neural network can describe the changes in drying shrinkage with time and compressive strength [18].
To verify the applicability of AAM age and compressive strength to the drying shrinkage prediction curve, the calculated values of drying shrinkage were compared with the test data. As shown in Figure 7(l), the prediction error is no more than 15%, meaning that the model prediction was in agreement with experimental data, thus substantiating that the model can predict the drying shrinkage of AAM. Figure 8 shows the effects of the compressive strength of AAM on drying shrinkage. It can be seen from Figure 8(a) that when the SF content is less than 10%, the slope of the curve will increase, indicating that SF reaction leads to water consumption, which causes the drying shrinkage value to rise. When the SF content is equal to 10%, the slope of the curve is the largest, indicating that SF not only causes water consumption at this time but also forms a low-density C-S-H gel that accelerates the rise of the drying shrinkage value.
When the SF content is greater than 10%, the SF may not fully react and the partially unreacted SF fills the voids, thereby reducing the increase in the shrinkage value. However, this affects the strength development, resulting in a decrease in the slope of the curve.
In Figure 8(b), the incorporation of NS helps to promote the reaction rate of the matrix, resulting in water consumption, but part of the stable gel produced makes the early strength development rate greater than the drying shrinkage evolution and the slope of the front part of the curve to be smaller. Above 1.5%, the slope decreases as the NS doping amount increases. When the doping amount of NS is 1.5%, the reaction leads to excessive water consumption, so that the drying shrinkage in the later period increases faster, and hence, the slope of the curve will increase. When the doping amount of NS is greater than 1.5%, its agglomeration will lead to a lower density of the matrix, resulting in a larger drying shrinkage value, which also affects the strength development, and hence, the slope of the curve increases.
From Figures 7 and 8, it can be deduced that the BP neural network can establish a prediction model based on age, compressive strength, and drying shrinkage value. And by adjusting the regression coefficients, the drying shrinkage prediction model can be formulated based on knowing the curing age and compressive strength.
us, the drying shrinkage value of modified alkali-activated mortar prepared using SF and NS with waste glass can be predicted with the following model: where z is drying shrinkage value, x is age, and y is compressive strength. e respective model fitting parameters and coefficients are listed in Table 4.
In actual field construction projects, it is usually difficult to measure the drying shrinkage of a structure. us, the compressive strength is often used as one of the detection and shrinkage representative measures. By measuring the age and compressive strength, drying shrinkage can be predicted in an engineering structure. ese predictions and assessments can be used to avoid large drying shrinkage related cracks in the structure and other hazardous failures.
In this paper, the prediction curves for age, compressive strength, and drying shrinkage established using BP neural network can predict the drying shrinkage of modified AAM made from SF and NS using waste glass powder-based additives. However, in actual engineering practice, its drying shrinkage is also affected by many interactive factors (such as water-to-cementitious material ratio, curing environment, 8 Advances in Civil Engineering  specimen size, and type of cementing material). Prior to shrinkage value verification, the proposed prediction model should not be considered to be universal-validation is still warranted. However, the establishment of this prediction model also provides reference datum for the practical application of AAM in engineering structures.

Conclusions and Recommendations
In this study, AAM prepared with different amounts of SFand NS-modified waste glass was used to investigate the compressive strength, flexural strength, ratio of bendingcompressive strength (Rt), and drying shrinkage properties along with a prediction model establishment. From the study results and findings, the following conclusions are drawn: (1) With an increase of the curing age, the compressive and flexural strengths of AAM increased, with the early strength development being faster than the later. e early drying shrinkage growth was also rapid, but decayed with age.
(2) Low content of SF can improve the flexural strength and toughness of the AAM as well as help to improve the matrix pore structure and, consequently, suppress the increase in drying shrinkage. When the amount of SF is greater than 10%, the strength of the AAM rapidly decays and adversely affects the Rt, which ultimately results in increased brittleness. erefore, the recommended optimum SF dosage is 5.0% for optimal performance.
(3) NS has a positive influence on the mechanical properties and toughness of AAM. When the NS dosage was 1.5%, it effectively suppressed the drying shrinkage. When the amount of NS was 2%, AAM had the highest compressive and flexural strengths, which inadvertently results in NS particle agglomeration, which not only increased the drying  shrinkage, but also adversely affected its strength development. erefore, the recommended optimum NS dosage is 1.5% for optimal performance. (4) rough regression analysis, a compressive strength prediction model of the AAM was generated. e relative error of the predicted value of the compressive strength and the experimental data did not exceed 5%, i.e., 95% prediction accuracy. erefore, the model was effectively reliable for SF and NS analysis as well as prediction of the AAM compressive strength. (5) Based on the BP neural network, a prediction model based on age, compressive strength, and drying shrinkage were mathematically established. e model exhibited good prediction accuracy and could reasonably predict the drying shrinkage of AAM containing SF or NS.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare that they have no conflicts of interest regarding the publication of this paper.