Data-driven model for ternary-blend concrete compressive strength prediction using machine learning approach
Introduction
Similar to every brittle material, the primary factor governing the compressive strength of concrete is the porosity. While there are other important characteristics of concrete, such as durability, impermeability, and volume stability, compressive strength has become universally accepted as the most important indication of concrete quality. Concrete compressive strength is affected by many factors, such as the quality of the raw materials (cement, supplementary cementitious materials, aggregates, water, etc.)., water-cement (w/c) ratio, coarse–fine aggregate ratio, age of concrete, proper vibration and compaction of concrete, relative humidity, and curing of concrete. The higher the w/c ratio, the greater the intergranular spacing between the cement particles and the greater the volume of voids left unfilled by hydration products after cement hydration. The coarse–fine aggregate (CA/FA) ratio is another influencing factor associated with the w/c ratio. The proportion of CA to FA influences the overall aggregate surface area, which affects the water demand of the concrete. With higher FA content relative to CA content, the water demand (w/c ratio) increases owing to the increased total aggregate surface area and vice versa if the FA content is lower. A higher or lower total aggregate surface area leads to a higher or lower w/c ratio, which in turn leads to a decrease or increase in the compressive strength, respectively. Since water is available for cement hydration, the compressive strength of concrete will continue, albeit at a continuously decreasing rate as it ages. Therefore, the age of concrete is synonymous with the degree of cement hydration provided that the concrete has not dried out or the surrounding temperature is not too low. In practical terms, a large portion of the concrete compressive strength would have developed after 28 d. To achieve a desirable compressive strength, curing is imperative for maximizing the concrete's potential for attainable compressive strength. Curing ensures the continuation of the hydration process in concrete, leading to continued strength gain. The hydration can continue due to the good preservation of moisture and temperature within the concrete for an adequate period of time.
Owing to the importance of the mechanical properties of concrete in reinforced concrete design codes and specifications, accurate prediction of these properties has become a concern. A significant concrete property is the compressive strength, which is usually measured after 28 days of standard curing [1]. Until recently, when computational intelligence, such as artificial neural networks (ANNs) [2], [3], [4], support vector regression (SVR) [5], [6], [7], [8], random forest (RF) [9], [10], [11], [12], etc., have been successfully applied in concrete research, estimating the mechanical properties of concrete has always been through linear and nonlinear regression methods [1], [13]. Other researchers have conducted studies to estimate the compressive strength using empirical methods where experimental data are calibrated, input ranges are limited, and constants required to describe the relationship between the inputs and the compressive strength are difficult to obtain [14], [15]. While some empirical models' effectiveness has been proven in some studies to save time and resources for future applications, multiple drawbacks remain in their development [16]. Several machine learning algorithms have been proposed as innovative predictive tools for predicting the compressive strength of concrete to overcome these weaknesses.
In concrete material studies, machine learning and data science have been used broadly in the optimization of concrete mixtures [17], [18], [19], fresh properties [20], [21], [22], [23], hardened properties [2], [5], [8], [10], [11], [12], [14], [24], [25], and durability properties [9], [26], [27], [28], [29] estimation of different types of concrete.
Using random forest, Silva et al. [30] presented a model to predict the compressive strength of concrete from 1030 experimental data with eight variables retrieved from Yeh's study [31]. The developed model outputs were compared with those from the support vector machine (SVM) and ANN. Comparing ANN and instance-based learning (IBL) or k-nearest neighbors (kNN), Dutta and Barai [32] explored the prediction efficiencies of these models in predicting the compressive strength of concrete using the mixture ingredients as features. Dao et al. [33] studied the sensitivity and robustness of Gaussian process regression (GPR) with five kernel functions and an ANN using a 500-run Monte Carlo simulation to estimate the compressive strength of high-performance concrete (HPC). The results revealed that the GPR with the Matern32 kernel function performed better than the ANN in predicting HPC compressive strength. Two models to forecast the compressive strength of HPC based on ANN ensembles (bagging and gradient boosting) with prediction accuracy enhancement through the coupling of discrete wavelength transform (DWT) with the ANN ensembles was proposed by Erdal et al. [34]. The DWT was concluded to be a useful tool for increasing the accuracy of the developed ANN ensembles.
Amid numerous other ML techniques, the least square support vector machine (LSSVM) has recently gained broader acceptance in concrete material science. [35], [36], [37], [38], [39]. With a careful review of available studies, it was found that there are few studies where LSSVM was adopted as the algorithm of choice for predicting the compressive strength of normal concrete. Xue [40] explored five different models and proposed a hybrid LSSVM model in one of these studies. Its hyperparameters were optimized to improve the model's estimation accuracy using improved particle swarm optimization (IMPSO) techniques. In another study, Aiyer et al. [41] explored the capability of the LSSVM and relevance vector machine (RVM) in the prediction of the compressive strength of SCC. While the RVM performed better, the margin of performance in comparison to LSSVM was minimal. Biswas et al. [35] also explored the excellent predictability of LSSVM in estimating concrete compressive strength in addition to other ML techniques. In another bid to estimate the concrete compressive strength accurately, Prayogo [42] developed a novel and advanced hybrid ML algorithm combining the robust prediction accuracy of LSSVM with a new metaheuristic, symbiotic organisms search (SOS).
The primary objective of our study is to develop a data-driven model with the ability to accurately predict the compressive strength of ternary blend (blast furnace slag, Portland cement, and fly ash) concrete using readily available experimental mixture parameters. This study proposes a least square support vector machine with coupled simulated annealing for parameter optimization (LSSVM-CSA) and genetic programming (GP) as predictive tools to solve the problem of estimating concrete compressive strength using experimentally acquired datasets. Our choice of LSSVM-CSA and GP as learning algorithms stems from their high prediction accuracy, successful deployment, and promising results recorded by many researchers [43], [44], [45], [46], [47], [48], [49], [50]. The parameters used in training the models include cement, water, coarse aggregate, fine aggregate, blast furnace slag, fly ash and superplasticizer, curing age as inputs, and compressive strength as output. In addition, the developed models were evaluated using statistical parameters such as the coefficient of determination (R2), root mean square error (RMSE), average relative deviation (ARD), and absolute average relative deviation (AARD) for the training and test datasets.
Section snippets
Genetic programming
Genetic programming (GP) is an example of Alan Turing's evolutionary algorithm in 1950 [51] and was developed by Koza [52], [53]. This method is closely related to the genetic algorithm developed based on the Darwinian natural selection principle [54], [55] and represents a symbolic regression model. GP and its other forms have been successfully applied to various research problems, including concrete formulation and design [43], [45], [46], [55], [56], [57], [58], [59]. The GP uses a heuristic
Data Description, data manipulation, and feature selection
The 1030 datasets for 17 studies used in constructing the data-driven model are the long-term compressive strength of ternary-blend concrete experimental data retrieved from the University of California dataset repository, Irvine [78], [79]. It is important to note that the compressive strength tests were conducted on cubic concrete specimens (150 mm) according to the laid-down requirements of ASTM C39 [80]. In the experimental study, the effects of a blend of cement, blast furnace slag, fly
Sensitivity analysis
The sensitivity of the LSSVM-CSA and GP models to all the independent variables used to develop the linear/nonlinear relationship between the input and output datasets was analyzed. The results are presented in Fig. 3. In other words, it may provide insight into the impact of input variables on the output variable [91]. A sensitivity analysis was performed to assess the effect of concrete mixture parameters (input variables) on the concrete compressive strength (target variable) using the
Conclusion
The objective of this study was to comparatively explore the predictive performance of LSSVM-CSA and GP in estimating the compressive strength of ternary-blend concrete using features from the datasets. The 1030 datasets used were retrieved from an experimental work kept in the UCI Machine Learning Repository (http://archive.ics.uci.edu/ml/). In this study, two data-driven machine learning models were developed and compared to optimize their prediction accuracy for concrete compressive strength
CRediT authorship contribution statement
Babatunde Abiodun Salami: Conceptualization, Methodology, Software, Investigation, Data Curation, Writing – original draft preparation, Writing – review & eiting, Visualization, Supervision, Project administration. Teslim Olayiwola: Methodology, Software, Validation, Resources, Writing – review & eiting, Validation. Tajudeen Adeyinka Oyehan: Resources, Project administration, Software, Writing – review & eiting. Ishaq A. Raji: Project administration, Writing – review & eiting.
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.
Acknowledgement
The authors gratefully acknowledge the support provided by the Center for Engineering Research, Research Institute, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia, for the reported work.
References (98)
- et al.
Prediction of the compressive strength of no-slump concrete: A comparative study of regression, neural network and ANFIS models
Constr. Build. Mater.
(2010) - et al.
Non-destructive prediction of concrete compressive strength using neural networks
Procedia Comput. Sci.
(2017) - et al.
Compressive strength prediction of environmentally friendly concrete using artificial neural networks
J. Build. Eng.
(2018) - et al.
Comparison of artificial neural network (ANN) and response surface methodology (RSM) prediction in compressive strength of recycled concrete aggregates
Constr. Build. Mater.
(2019) - et al.
Modelling the strength of lightweight foamed concrete using support vector machine (SVM)
Case Stud. Constr. Mater.
(2017) - et al.
Ensemble machine learning model for corrosion initiation time estimation of embedded steel reinforced self-compacting concrete
Measurement
(2020) - et al.
A generalized method to predict the compressive strength of high-performance concrete by improved random forest algorithm
Constr. Build. Mater.
(2019) - et al.
Modelling uniaxial compressive strength of lightweight self-compacting concrete using random forest regression
Constr. Build. Mater.
(2019) - et al.
A comparison of machine learning methods for predicting the compressive strength of field-placed concrete
Constr. Build. Mater.
(2019) - et al.
Machine learning prediction of mechanical properties of concrete: Critical review
Constr. Build. Mater.
(2020)
Computational design optimization of concrete mixtures: A review
Cem. Concr. Res.
Multi-objective optimization of concrete mixture proportions using machine learning and metaheuristic algorithms
Constr. Build. Mater.
Linear and nonlinear SVM prediction for fresh properties and compressive strength of high volume fly ash self-compacting concrete
Constr. Build. Mater.
Machine learning to predict properties of fresh and hardened alkali-activated concrete
Cem. Concr. Compos.
prediction of interface yield stress and plastic viscosity of fresh concrete using a hybrid machine learning approach
Adv. Eng. Inf.
Modelling the fresh properties of self-compacting concrete using support vector machine approach
Constr. Build. Mater.
Predicting compressive strength of lightweight foamed concrete using extreme learning machine model
Adv. Eng. Softw.
Predicting the time to corrosion initiation in reinforced concrete structures exposed to chlorides
Cem. Concr. Res.
Modeling the corrosion initiation time of slag concrete using the artificial neural network
HBRC J.
Predicting concrete corrosion of sewers using artificial neural network
Water Res.
High performance concrete compressive strength forecasting using ensemble models based on discrete wavelet transform
Eng. Appl. Artif. Intell.
Toward genetic programming (GP) approach for estimation of hydrocarbon/water interfacial tension
J. Mol. Liq.
A new empirical model for estimation of crude oil/brine interfacial tension using genetic programming approach
J. Petrol. Sci. Eng.
A simple correlation to predict surface tension of binary mixtures containing ionic liquids
J. Mol. Liq.
Prediction of Cementation Factor for Low-Permeability Iranian Carbonate Reservoirs Using Particle Swarm Optimization-Artificial Neural Network Model and Genetic Programming Algorithm
J. Petrol. Sci. Eng.
A data-driven model for predicting the effect of temperature on oil-water relative permeability
Fuel
Modeling the acentric factor of binary and ternary mixtures of ionic liquids using advanced intelligent systems
Fluid Phase Equilib.
Determination of bubble point pressure and oil formation volume factor: Extra trees compared with LSSVM-CSA hybrid and ANFIS models
Fuel
New correlations for predicting pure and impure natural gas viscosity
J. Nat. Gas Sci. Eng.
Genetic programming approach for prediction of compressive strength of concretes containing rice husk ash
Constr. Build. Mater.
Application of Artificial Intelligence-based predictive methods in Ionic liquid studies: A review
Fluid Phase Equilib.
Rutting depth prediction of hot mix asphalts modified with forta fiber using artificial neural networks and genetic programming technique
Constr. Build. Mater.
Analytical and genetic programming model of compressive strength of eco concretes by NDT according to curing temperature
Constr. Build. Mater.
Empirical modelling of shear strength of RC deep beams by genetic programming
Comput. Struct.
A comprehensive survey on support vector machine classification: Applications, challenges and trends
Neurocomputing
A support vector machine algorithm to classify lithofacies and model permeability in heterogeneous reservoirs
Eng. Geol.
Weighted least squares support vector machines: robustness and sparse approximation
Neurocomputing
High-performance concrete compressive strength prediction using time-weighted evolutionary fuzzy support vector machines inference model
Autom. Constr.
Compressive strength prediction of high-performance concrete using gradient tree boosting machine
Constr. Build. Mater.
Evolving LSSVM and ELM models to predict solubility of non-hydrocarbon gases in aqueous electrolyte systems
Measurement
Connectionist intelligent model estimates output power and torque of stirling engine
Renew. Sustain. Energy Rev.
prediction of automotive engine power and torque using least squares support vector machines and Bayesian inference
Eng. Appl. Artif. Intell.
Utilization of a least square support vector machine (LSSVM) for slope stability analysis
Sci. Iran.
Dataset of long-term compressive strength of concrete with manufactured sand
Data Brief.
Experimental study on long-term compressive strength of concrete with manufactured sand
Constr. Build. Mater.
Compressive strength and sulfate resistance properties of concretes containing Class F and Class C fly ashes
Constr. Build. Mater.
Performance characteristics of high-volume Class F fly ash concrete
Cem. Concr. Res.
Influence of fly ash on corrosion resistance and chloride ion permeability of concrete
Constr. Build. Mater.
Impact of added water and superplasticizer on early compressive strength of selected mixtures of palm oil fuel ash-based engineered geopolymer composites
Constr. Build. Mater.
Cited by (45)
New-generation machine learning models as prediction tools for modeling interfacial tension of hydrogen-brine system
2024, International Journal of Hydrogen EnergyHybrid random forest-based models for predicting shear strength of structural surfaces based on surface morphology parameters and metaheuristic algorithms
2023, Construction and Building MaterialsBuilding energy loads prediction using bayesian-based metaheuristic optimized-explainable tree-based model
2023, Case Studies in Construction MaterialsCompressive strength prediction of ternary-blended concrete using deep neural network with tuned hyperparameters
2023, Journal of Building Engineering