Machine learning modeling integrating experimental analysis for predicting the properties of sugarcane bagasse ash concrete

https://doi.org/10.1016/j.conbuildmat.2021.125634Get rights and content

Highlights

  • Machine learning models were used to predict strength of bagasse ash concrete.

  • Waste sugarcane bagasse ash (SCBA) was processed, characterized and optimized.

  • A number of concrete samples were prepared in laboratory to validate the ML model.

  • Robust models were developed using integration of literature and experimental data.

Abstract

The present study aimed to develop models for estimation of the compressive strength (fc) of sugarcane bagasse ash (SCBA) concrete through experimental testing and three machine learning (ML) approaches, namely gene expression programming (GEP), random forest regression (RFR) and support vector machine (SVM). The models were calibrated based on a widespread literature dataset of five input variables i.e., SCBA dosage (SCBA%), the quantity of fine aggregate (FA) and coarse aggregate (CA), water-cement ratio (W/C) and cement content (CC). The performance of each model was evaluated using relative squared error (RSE), root mean square logarithmic error (RMSLE), root mean squared error (RMSE), mean absolute error (MAE), Nash-Sutcliffe efficiency (NSE), percentage of relative root mean squared error (RRMSE%), coefficient of determination (R2) and performance index (PI). The suggested models prediction was checked and validated against the actual dataset acquired from laboratory tested SCBA concrete. The comparative study of the models revealed that RFR is an effective approach providing a strong correlation between actual and estimated outcome. The R2 and NSE for all the models are above 0.85 each with RRMSE% and PI less than 10% and 0.2, respectively. The GEP leads in providing a simple and reliable mathematical expression for estimation of fc of SCBA concrete. The sensitivity analysis reflected the increasing order of the influence of input variables followed the trend: CC (55.73%) > W/C (17.15%) > CA (16.38%) > SCBA% (6.38%) > FA (3.76%) and are in close agreement with the experimental study conducted. Each proposed ML approach outperform on the literature data as well as dataset from laboratory tested SCBA concrete and also having superior generalization and prediction capacity. Conclusively, the results of this research can enable practitioners, researchers, and designers in quickly evaluating the f'c of SCBA concrete, thereby reducing environmental susceptibilities and resulting in safer, faster, and more sustainable construction from the perspective of eco-friendly waste management.

Introduction

The planet earth is facing critical issues such as climate change and resulting snow/glacier melt due to massive CO2 emission which is mainly caused by anthropogenic activities [1]. The industrial processes are considered to have adverse environmental impacts with the construction industry alone responsible for 50% of global greenhouse gas (GHG) emissions [2], [3]. The Portland cement, a major constituent of concrete, has a substantial role in the GHG emissions [4], [5]. The production of cement produces 7% of CO2 in the environment [6], [7]. Moreover, the calcination process of cement contributes approximately 50% out of 7% CO2 emitted into the atmosphere and the remaining 50% of the energy used in the cement manufacturing process [8], [9]. Moreover, the annual demand of cement production is almost 4000 million tons and the annual use of cement around the globe will hit 6000 million tons by 2060 [10], [11]. Therefore, it is indispensable to adopt new practices to search for alternative binding and waste materials for concrete production which requires less energy with minimal environmental pollution. The use of waste materials in the construction industry not only give the utmost effect but reduces the CO2 emission and disposal issues.

Globally, an adamant amount of wastes are generated because of rapid industrialization and has malignant effects on the environment [12], [13]. Meanwhile, research is focused on condensing the underlying adverse impacts on the environment by utilizing wastes and by-products of industries for concrete production. Incorporation of these wastes produces viable, economical and durable concrete due to low-carbon content and substitute to primary binding material in concrete [14], [15]. Therefore, green concrete is gaining enormous attention in the construction sector [16], [17]. One of the potential agricultural wastes from the sugar industry is sugarcane bagasse ash (SCBA) and could be used as a potential alternative material to make sustainable concrete [18]. The addition of SCBA in concrete not only minimizes the disposal issues in the land but preserves the natural environment as well [12].

Concrete ultimate aspect is the compressive strength that depends on various factors used in its production i.e. concrete ingredients, mix design, material type, and testing methods [19]. Also, the dosages of additives and curing practices affect the mechanical properties of concrete [20], [21]. Indeed, proper knowledge about the interaction between parameters and strength is needed. Laboratory experimental work is quite a time-consuming process and requires an adamant amount of resources to assess the strength of concrete. In contrast, the modeling tools can be easily used for the purpose to condense the practical work for the valuation of concrete properties [22]. However, a limited amount of test records and constraints in the models parametric ranges reduces trust in the modeling tools. Literature study demonstrated that a large database is required to make an accurate model that predicts the mechanical response of concrete using generated equations by models. The development of an effective model leads to time-saving with an accurate prediction of strength [12], [23]. Various researchers have used regression based models for the assessment of concrete compressive strength [24], [25], [26]. Some prominent shortcomings were witnessed in regression-based modeling. The regression analysis is based on some pre-defined linear or non-linear equations and pre-assume the residual normality, therefore, restricting its wide scale application [27], [28]. The accuracy and performance of regression models declines during validation stage [29]. Furthermore, the available equations in codes are facing serious concerns regarding the compressive strength prediction as these equations are created without considering the amount of supplementary cementitious materials [28].

Given the above shortcomings of the regression based techniques, research has now been focused on using the prominent features of machine learning (ML) techniques. Artificial neural network (ANN) is a frequently used technique for evaluating concrete mechanical properties [30], [31], [32]. The ANN could not provide the information regarding the adopted principles for modeling, therefore, it is deliberated as a black-box model. The neural network formulation is too multifaceted and unable to provide mathematical expressions [12], [23]. The support vector machine (SVM) is another ML technique, which can design and efficiently deal with non-linear regression problems. SVM showed a high generalization capability and is advantageous in reaching improved global optimum results than local optimal [33]. The ensemble learning methods like decision tree (DT) and random forest (RF) can also be used to predict the response parameter using tree-like structures. DT utilizes the whole dataset with the variables of interest, while RF randomly chooses the significant factors inside the parameters and develops several ensemble learning trees in predictions [34]. Then, the averaged values of these predictions is estimated and set with majority votes provides an accurate estimate [34]. Both, SVM and RF can be only used as a predictor and does not give a simple mathematical equation. The strong structural design of SVM and RF algorithms enables researchers to better predict the results. Moreover, the advancement of these ML approaches have broadened the uses of these models to a variety of fields due to its capability to overcome challenging and complicated problems with high accuracy and precision. Chou et al. (2014) [33] investigated different ML methods like ANN and SVM to forecast the compressive loading behavior of concrete. Jalal et al. (2020) [35] extended the study of the SVM method to a further complex situation and evaluated the strength of concrete mixes incorporating waste rubber tire. Sañudo, R. et al. (2016) [36] used RF approach for forecasting the strength and examined the importance and impact of input variables. Farooq et al. (2020) [37] used ANN, GEP, RF and DT and predicted the compression capacity of high strength concrete with R2 equals to 0.89, 0.90, 0.96, and 0.90, respectively. The modified genetic programming (GP) technique called gene expression programming (GEP) is considered an excellent technique, as it deliver a prediction equation for future use on unseen data [19]. The GEP algorithm can model complex programs and provide a mathematical expression that can be used to get an accurate output. The applications of the GEP algorithm have been extended to the Civil engineering domain for modeling and solving complex issues [27], [38], [39], [40]. The GEP was used by several researchers to model and investigate the properties of concrete. Abdollahzadeh et al. (2017) [19] used two GEP models to forecast the strength of concrete. The authors reported the high prediction capability of the GEP technique to model compressive strength of concrete. Binici et al. (2009) [41] developed two models for the heat of hydration of blended and Portland cement. The parameters including cement fineness, blast furnace slag replacement ratio, basaltic pumice replacement ratio, grinding type, clinker/gypsum ratio, and time were used as inputs in the first model. While in the second model, cement fineness, grinding type, and time was used as an inputs. The GEP models generate the R2 value from 0.96 to 0.98. The authors reported that GEP could be used as a powerful approach for modeling and prediction. Sarıdemir and Billir (2016) [20] used literature data to model the modulus of elasticity of concrete with fly ash and the authors reported an excellent performance and high accuracy of GEP method.

The aforementioned discussion highlighted that different ML models have been effectively used and developed to predict the properties of various types of concrete. Literature study also revealed that ML models such as supervised, unsupervised and ensemble based models behaves differently on a given dataset. This leads the authors to use supervised as well as ensemble machine learning algorithm in forecasting the compressive strength of SCBA concrete using model-based empirical relation. The development of ML techniques and their comparative study made it very simple to choose best model to cater the difficulties in determining the concrete properties [12]. Therefore, the aim of the present study is to develop different ML models ranging from supervised to ensemble techniques and then integrate these ML models with the dataset of experimental results for the prediction of SCBA concrete properties. This study makes various contribution to literature as follows;

  • An effective and reliable models for the compressive strength (f’c) of SCBA concrete using three different ML approaches, namely gene expression programming (GEP), random forest regression (RFR), and support vector machine (SVM) were developed in accordance with the data acquired from available literature. Five different variables i.e., SCBA dosage (SCBA%), the quantity of fine aggregate (FA) and coarse aggregate (CA), water-cement ratio (W/C), and cement content (CC) were considered as an effective input variables.

  • Thereafter, the waste SCBA was processed employing sieving (75 µm) and ball mill grinding, characterized by mean of x-ray diffraction (XRD), x-ray fluorescence (XRF), and scanning electron microscopy (SEM). Concrete cylinders were then prepared in the lab using the processed SCBA in different proportions i.e. 10, 20, 30 and 40% by partially replacing the cement to determine the compressive strength.

  • The dataset of the results obtained from laboratory tested SCBA concrete were used to validate and assess the behavior of the aforementioned models for SCBA concrete.

The adopted modeling techniques and the outcome from the validated model through laboratory-derived dataset can enable practitioners, researchers, and designers in quickly evaluating the f'c of SCBA concrete, thereby reducing environmental susceptibilities and resulting in safer, faster, and more sustainable construction from the perspective of eco-friendly waste management.

Section snippets

Gene expression programming (GEP)

J. Holland [42] was the first who introduced the genetic algorithm (GA) and got inspired by Darwin's evolutionary theory. The GAs exemplifies the biological process and the solution is highlighted in the form of chromosomes. Genetic programming (GP) is a technique that creates genetic evaluation based models and adopted the properties of regression and neural technique. The GP technique works on the Darwinian reproduction principle simplified in a computer-based solution.

The adapted GP version

Materials and optimization

Sugarcane bagasse (SCB) was collected from a small sugar industry which is located in Malakand Khyber Pakhtunkhwa, Pakistan. The obtained SCB was in wet form as it is residual waste which was then placed in an open atmosphere to dry as shown in Fig. 2a. The dry SCB was then burned in an open atmosphere to get sugarcane bagasse ash (SCBA) as presented in Fig. 2b. The open burned SCBA was light black in color which indicates carbon content mainly due to incomplete burning [79]. The SCBA was

Properties of fresh concrete

Density and slump of different concrete mixes were carried out with SCBA as a partial replacement of cement by varying dosages (0 to 40) %. The effect of the addition of SCBA in term of slump test and fresh concrete density is illustrated in Fig. 6a and 6b, respectively. The slump values of SCBA concrete increases as compared to the control one (CM). The increase in workability may be attributed to the fact that the SCBA was sieved from 75 µm as the unburnt particles in SCBA rises the water

Conclusion

The present research adopted a unique approach where ML based approaches were integrated with experimental investigation for assessing the compressive strength (f’c) of sugarcane bagasse ash concrete. In the first step, three ML techniques including gene expression programming (GEP), random forest regression (RFR) and support vector machine (SVM) were developed using five input variables i.e., cement content (CC), water-to-cement ratio (W/C), percentage of bagasse ash (SCBA%), quantity of fine

Funding

This research received no external funding.

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.

References (90)

  • L. Chen et al.

    Sustainable stabilization/solidification of municipal solid waste incinerator fly ash by incorporation of green materials

    J. Cleaner Prod.

    (2019)
  • L. Wang et al.

    Novel synergy of Si-rich minerals and reactive MgO for stabilisation/solidification of contaminated sediment

    J. Hazard. Mater.

    (2019)
  • H. Du et al.

    Properties of high volume glass powder concrete

    Cem. Concr. Compos.

    (2017)
  • K.H. Tan et al.

    Use of waste glass as sand in mortar: Part I-Fresh, mechanical and durability properties

    Cem. Concr. Compos.

    (2013)
  • A. Bahurudeen et al.

    Performance evaluation of sugarcane bagasse ash blended cement in concrete

    Cement and Concrete Composites

    (2015)
  • A. Kliszczewicz et al.

    Differences in instantaneous deformability of HS/HPC according to the kind of coarse aggregate

    Cem. Concr. Compos.

    (2002)
  • J. Zhang et al.

    Experimental Investigation and Prediction of Compressive Strength of Ultra-High Performance Concrete Containing Supplementary Cementitious Materials

    Adv. Mater. Sci. Eng.

    (2017)
  • S.M. Mousavi et al.

    A new predictive model for compressive strength of HPC using gene expression programming

    Adv. Eng. Softw.

    (2012)
  • I.-C. Yeh

    Modeling of strength of high-performance concrete using artificial neural networks

    Cem. Concr. Res.

    (1998)
  • F.E. Jalal et al.

    Predictive modeling of swell-strength of expansive soils using artificial intelligence approaches: ANN, ANFIS and GEP

    J. Environ. Manage.

    (2021)
  • I.-C. Yeh

    Modeling slump flow of concrete using second-order regressions and artificial neural networks

    Cem. Concr. Compos.

    (2007)
  • T. Ji et al.

    A concrete mix proportion design algorithm based on artificial neural networks

    Cem. Concr. Res.

    (2006)
  • J.-S. Chou et al.

    Machine learning in concrete strength simulations: Multi-nation data analytics

    Constr. Build. Mater.

    (2014)
  • F. Farooq et al.

    Predictive modeling for sustainable high-performance concrete from industrial wastes: A comparison and optimization of models using ensemble learners

    J. Cleaner Prod.

    (2021)
  • M. Jalal et al.

    Behavior assessment, regression analysis and support vector machine (SVM) modeling of waste tire rubberized concrete

    J. Cleaner Prod.

    (2020)
  • R. Sañudo et al.

    Track transitions in railways: A review

    Constr. Build. Mater.

    (2016)
  • S. Mahdinia et al.

    Effect of cement strength class on the prediction of compressive strength of cement mortar using GEP method

    Constr. Build. Mater.

    (2019)
  • M. Azimi-Pour et al.

    ANN and GEP prediction for simultaneous effect of nano and micro silica on the compressive and flexural strength of cement mortar

    Constr. Build. Mater.

    (2018)
  • M.I. Shah et al.

    Environmental assessment based surface water quality prediction using hyper-parameter optimized machine learning models based on consistent big data

    Process Saf. Environ. Prot.

    (2021)
  • I. Azim et al.

    Prediction model for compressive arch action capacity of RC frame structures under column removal scenario using gene expression programming

    Structures

    (2020)
  • A.M. Abd et al.

    Modelling the strength of lightweight foamed concrete using support vector machine (SVM)

    Case Stud. Constr. Mater.

    (2017)
  • K. Ganesan et al.

    Evaluation of bagasse ash as supplementary cementitious material

    Cem. Concr. Compos.

    (2007)
  • A. Rerkpiboon et al.

    Strength, chloride resistance, and expansion of concretes containing ground bagasse ash

    Constr. Build. Mater.

    (2015)
  • S. Rukzon et al.

    Utilization of bagasse ash in high-strength concrete

    Mater. Des.

    (2012)
  • G.C. Cordeiro et al.

    Ultrafine grinding of sugar cane bagasse ash for application as pozzolanic admixture in concrete

    Cem. Concr. Res.

    (2009)
  • P. Jagadesh et al.

    Evaluation of mechanical properties of Sugar Cane Bagasse Ash concrete

    Constr. Build. Mater.

    (2018)
  • M. Jahanzaib Khalil et al.

    Utilization of sugarcane bagasse ash as cement replacement for the production of sustainable concrete–A review

    Constr. Build. Mater.

    (2021)
  • M.F. Iqbal et al.

    Prediction of mechanical properties of green concrete incorporating waste foundry sand based on gene expression programming

    J. Hazard. Mater.

    (2020)
  • M.A. Getahun et al.

    Artificial neural network based modelling approach for strength prediction of concrete incorporating agricultural and construction wastes

    Constr. Build. Mater.

    (2018)
  • A.N. Mandeville et al.

    River flow forecasting through conceptual models part III-The Ray catchment at Grendon Underwood

    J. Hydrol.

    (1970)
  • M.F. Iqbal et al.

    Sustainable utilization of foundry waste: Forecasting mechanical properties of foundry sand based concrete using multi-expression programming

    Sci. Total Environ.

    (2021)
  • V.D. Katare et al.

    Experimental characterization of sugarcane biomass ash–A review

    Constr. Build. Mater.

    (2017)
  • A. Bahurudeen et al.

    Influence of different processing methods on the pozzolanic performance of sugarcane bagasse ash

    Cem. Concr. Compos.

    (2015)
  • T. Akram et al.

    Production of low cost self compacting concrete using bagasse ash

    Constr. Build. Mater.

    (2009)
  • A.H. Gandomi et al.

    Assessment of artificial neural network and genetic programming as predictive tools

    Adv. Eng. Softw.

    (2015)
  • Cited by (44)

    View all citing articles on Scopus
    View full text