Elsevier

Measurement

Volume 113, January 2018, Pages 99-107
Measurement

New prediction models for unconfined compressive strength of geopolymer stabilized soil using multi-gen genetic programming

https://doi.org/10.1016/j.measurement.2017.08.043Get rights and content

Highlights

  • Two MGGP models are proposed to predict the UCS of geopolymer stabilized soil.

  • The MGGP models perform superior to MLSR models.

  • The contribution of each parameter is evaluated through a sensitivity analysis.

  • The effect of each parameter on the output in parametric study is confirmed with soil mechanics concepts.

Abstract

This study presents new models for the prediction of unconfined compressive strength (UCS) of geopolymer stabilized clayey soils using a modified branch of genetic programming, called multi-gen genetic programming (MGGP). The proposed MGGP models incorporate several parameters affecting the behavior of the UCS of the clayey stabilized soil. UCS is formulated in terms of percentages of fly ash, ground granulated blast furnace slag, liquid limit, plastic limit, plasticity index, molar concentration, alkali to binder ratio, and ratios of sodium and silicon to aluminum. The importance of each predictor variable is measured through a sensitivity analysis. The validity of the models and the trend of the results are verified by performing parametric study. The parametric study results are also in good agreement with previous studies. The results indicate that the proposed equations are capable of evaluating UCS accurately.

Introduction

In the last few decades, significant attention has been devoted to the analysis of contribution of clay to structural damage [56]. The sensitivity of clay to climate changes and its collapse and swelling due to water content changes is the main reason for the resultant damages. Since 1950s, various methods have been introduced for reducing destructive effect of expansive clays. Some of the commonly-used methods are mechanical stabilization (e.g. compaction [14]) and altering the physicochemical properties of the soil though addition of chemical agents or process of ion changing [26], [55].

Lime and cement injections were the most widely used methods for chemical stabilization of clay. However, hydration starts when water and cement are mixed and hydrated calcium silicates (CSH), aluminates (CHA) and lime (Ca(OH)2) are generated. This causes a rise in pore water PH. In 1971, Carroll et al. [8] found the reactivity of clay minerals with acids and alkalines and afterward the usability of chemicals other than lime and cement was investigated. Also, in order to reduce the stabilization costs, replacing cement with low cost waste materials has been studied [24]. The effect of Portland cement (PC) manufacturing on the amount of greenhouse released in the atmosphere and consequently, global warming has been another reason for finding a new agent with low carbon dioxide release [53]. Therefore, utilization of geopolymers or alkali-activated aluminosilicate cements has increased over the past few decades due to their high strength, durability and eco-friendly characteristics [13]. These materials demonstrate ceramic-like behaviors (e.g. resistance against high temperature and acid) because of their hardening process [53].

The basic material for geopolymers comprises high amount of Silicon (Si) and Aluminum (Al). In the process of geopolymerization, this material should be activated with alkaline liquid to produce an amorphous polymeric material [21]. Sodium silicate, sodium hydroxide, potassium hydroxide and their combinations are some of the common alkaline activators used for this process [29]. The atomic ratio of Sodium to Aluminum (Na/Al) and Silicon to Aluminum (Si/Al) in the mixture controls the reaction kinetics of geopolymer synthesis in the process. Therefore, these ratios influence the ultimate strength of the product [15], [21], [29], [40]. It should be noted that geopolymers of various aluminosilicate sources have different physical, mechanical, chemical, thermal and microstructure properties, but similar macroscopic characteristics [15].

So far, there are a few studies carried out on using geopolymer binder as a soil stabilizer [33]. Two of the most common geopolymers used for stabilization of soil are fly-ash (FA) and ground granulated blast furnace slag (GGBS). FA is a product of burning pulverized coal in electric power plants. On the other hand, ground granulated blast furnace slag is a product of the blast-furnaces used for making iron [48]. Since FA does not possess any plasticity, the plasticity index (PI) of the mixture of clay-FA reduces by increasing the amount of FA. Also, by colloidal reaction, it can increase the workability of the mixture [7]. More importantly, although FA barely possesses cementing properties, it reacts with calcium oxide in presence of water and produces highly cementations water insoluble products (pozzolanic reactivity). GGBS, if used without any activator, demonstrates slow hydration rate and low early strength. Therefore, chemical activation is usually used for the acceleration of the hydration process of GGBS [20]. Lime, PC, alkalis and magnesium are some of the most common activators for GGBS. In recent decades, the application of FA and GGBS as clay stabilizers has been demonstrated to be effective [9], [49], [53].

In the most recent researches on the effect of FA on clay stabilization [7], [11], [12], [27], [36], it is found that the unconfined compressive strength UCS) of clayey soil increases by addition of FA. Kampala et al. [27] demonstrated that the optimal content of FA was about 20% while the trend of UCS afterward was almost constant. In this context, Bose [7] showed a decreasing trend for UCS. Horpibulsuk et al. [23] reported similar results for a mixture of calcium carbide residue (CCR) and FA, with small differences due to the amount of CCR. They also mentioned that their mixture is more useful than PC in soil stabilization.

In contrary to FA, studies on binder content, ratio of alkaline to binder and molar strength of alkali affecting mechanical properties of GGBS are limited [33]. However, previous studies show improvements in USC of clayey soil in presence of GGBS [6], [9], [20], [43], [57]. Also, the UCS values can be different based on the type of the activator used for GGBS [20], [57]. In a comparison with PC mixes, Yi et al. [57] found up to 4 times higher 28-day UCS values for a GGBS-MgO mix.

As mentioned earlier, several parameters affect the UCS. Measuring the UCS of soil samples need cumbersome and time consuming lab efforts. To cope with this issue, developing predictive models can be beneficial. For developing such behavioral models, several methods can be employed such as traditional regression techniques [38], [39]. However, the regression techniques have some major drawbacks like assuming pre-specified linear or nonlinear relationships between the inputs and output, which is not always valid [16]. An alternative method is artificial neural network (ANN) which has been implemented in several engineering applications [1], [2], [3], [4], [25], [31], [32], [51], [54], [58]. One of the most recent studies for developing a predictive model to calculate UCS of geopolymer stabilized clayey soils using ANN is done by Mozumder and Laskar [33]. Despite the acceptable performance of the ANN models, they do not provide practical equations to calculate the results. To resolve this issue, genetic programming (GP) seems to be an efficient alternative. GP is a machine learning technique that generates computer programs relating the inputs to outputs.

In this study, a modified type of GP, called multi-gen genetic programming (MGGP), is utilized to develop models for predicting UCS of geopolymer stabilized clayey soils. The models consider plasticity index, molar concentration, alkali to binder ratio, percent of FA and GGBS and the ratios of Na/Al and Si/Al as the input variables. The performance of the models is assessed and compared with traditional regression and ANN models. Moreover, the effect of each parameter on the value of UCS is studied by performing parametric study.

Section snippets

Genetic programming

GP is an evolutionary computing method that uses the principle of Darwinian natural selection in order to create computer programs to solve a problem. In 1992, Koza [30] introduced GP as a branch of genetic algorithm. GP covers a high level of diversity by breeding random different computer programs as the population. The programs are structured like trees containing functions and terminals. Two sets should be defined as the sources of functions and terminals. If they are rich enough, tree

Prediction of UCS by development of MGGP models

In this study, MGGP was implemented to find the relationships between the UCS of geopolymer stabilized clayey soil and parameters affecting it. The final predictor parameters here were selected based on a literature review [31], [34], [41], [49] and trial and error. The UCS prediction model is considered to be as follows:UCS=f(LL,PL,PI,%S,%FA,M,A/B,Na/Al,Si/Al)where,

  • LL = Liquid limit

  • PL = Plastic limit

  • PI = Plasticity index

  • %S = Percentage of ground granular blast furnace slag

  • %FA = Percentages of fly ash

  • M = 

Performance analysis

According to a logical hypothesis [28], [50], a strong correlation exists between predicted and measured values if R value of the model is more than 0.8 and the error values (e.g. MAE and MSE) are at minimum. Low error values demonstrate the predictive ability of the model. In addition, it is important that the error values of both training and testing datasets be similar to each other. This similarity is an indication of the generalized performance of the model [35]. Table 3 shows a detailed

Parametric study

In order to investigate the effect of each parameter on the predicted values, a parametric study was performed. In this procedure, a predictive variable should be changed within its range at a time while the other variables are constant at their average values. The robustness of the equations can determined by investigating how well the predicted values agree with the essential physical behavior of the problem [31]. The propensity of UCS to each of the parameters was investigated and the

Conclusions

In this study, the MGGP method was implemented to assess the UCS of the geopolymer stabilized clayey soils. Two different equations were derived using different combinations of the effective parameters. The final parameters were LL, PL, PI, S, FA, M, A/B, Na/Al and Si/Al, which were selected based on a literature review. Also, a comprehensive database was used for developing of the models. The MGGP-based models give authentic predictions of UCS. Moreover, they performed superior to the

References (58)

  • F. Puertas et al.

    Alkali-activated fly ash/slag cements: strength behaviour and hydration products

    Cem. Concr. Res.

    (2000)
  • A. Sathonsaowaphak et al.

    Workability and strength of lignite bottom ash geopolymer mortar

    J. Hazard. Mater.

    (2009)
  • K. Somna et al.

    NaOH-activated ground fly ash geopolymer cured at ambient temperature

    Fuel

    (2011)
  • P. Sukmak et al.

    Strength development in clay–fly ash geopolymer

    Constr. Build. Mater.

    (2013)
  • W. Zhang et al.

    Multivariate adaptive regression splines and neural network models for prediction of pile drivability

    Geosci. Front.

    (2016)
  • A.H. Alavi et al.

    Damage detection using self-powered wireless sensor data: an evolutionary approach

    Measurement

    (2016)
  • G. Alexander et al.

    Beware of Q2

    J. Mol. Graph Model

    (2002)
  • Y.-l. Bo et al.

    Effect of wetting-drying cycles on strength properties of ground granulated blast-furnace slag and magnesium oxide-stabilized soft clay

    Chin. J. Geotech. Eng.

    (2013)
  • B. Bose

    Geo engineering properties of expansive soil stabilized with fly ash

    Electron. J. Geotech. Eng.

    (2012)
  • D. Carroll et al.

    Reactivity of clay minerals with acids and alkalies

    Clays Clay Miner

    (1971)
  • L. Chen et al.

    The Effects of GGBS on the Engineering Properties of Stabilized Soil

    Chin. J. Underground Space Eng.

    (2013)
  • N.C. Consoli et al.

    Key parameters for strength control of artificially cemented soils

    J. Geotech. Geoenviron. Eng.

    (2007)
  • N. Cristelo et al.

    Deep soft soil improvement by alkaline activation

    Proc. ICE-Ground Improvement

    (2011)
  • J. Davidovits

    Geopolymers

    J. Therm. Anal.

    (1991)
  • L. Dubose

    Compaction control solves heaving clay problems

    Civil Eng.

    (1955)
  • P. Duxson et al.

    Geopolymer technology: the current state of the art

    J. Mater. Sci.

    (2007)
  • A.H. Gandomi et al.

    A new multi-gene genetic programming approach to non-linear system modeling. Part II: geotechnical and earthquake engineering problems

    Neural Comput. Appl.

    (2012)
  • A.H. Gandomi et al.

    A new multi-gene genetic programming approach to nonlinear system modeling. Part I: materials and structural engineering problems

    Neural Comput. Appl.

    (2012)
  • A.H. Gandomi et al.

    Nonlinear genetic-based models for prediction of flow number of asphalt mixtures

    J. Mater. Civ. Eng.

    (2010)
  • Cited by (63)

    • Ultrasonic assessment of cement-stabilized soils: Deep learning experimental results

      2023, Measurement: Journal of the International Measurement Confederation
    View all citing articles on Scopus
    View full text