Elsevier

Construction and Building Materials

Volume 180, 20 August 2018, Pages 320-333
Construction and Building Materials

A modified firefly algorithm-artificial neural network expert system for predicting compressive and tensile strength of high-performance concrete

https://doi.org/10.1016/j.conbuildmat.2018.05.201Get rights and content

Highlights

Abstract

The compressive and tensile strength of high-performance concrete (HPC) is a highly nonlinear function of its constituents. The significance of expert frameworks for predicting the compressive and tensile strength of HPC is greatly distinguished in material technology. This study aims to develop an expert system based on the artificial neural network (ANN) model in association with a modified firefly algorithm (MFA). The ANN model is constructed from experimental data while MFA is used to optimize a set of initial weights and biases of ANN to improve the accuracy of this artificial intelligence technique. The accuracy of the proposed expert system is validated by comparing obtained results with those from the literature. The result indicates that the MFA-ANN hybrid system can obtain a better prediction of the high-performance concrete properties. The MFA-ANN is also much faster at solving problems. Therefore, the proposed approach can provide an efficient and accurate tool to predict and design HPC.

Introduction

Nowadays, high-performance concrete (HPC) has many applications in civil engineering, including high-rise buildings, high-speed railways, bridges and extreme loading (e.g., fire, blast, impact) resistance systems [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12]. HPC has not only high compressive strength but also low permeability, and a high modulus of elasticity. Compared with ordinary concrete, which is composed of three main components including water, fine and coarse aggregates, and cement, HPC is supplemented by an additional cementitious material, for instance, silica fume, nano-silica, blast furnace slag and fly ash to enhance its compressive strength [13], [14], [15]. However, the properties of HPC depend on many elements such as mix proportions, material quality and the age of concrete [16].

Therefore, predicting the compressive and tensile strength of HPC is an important task because it can help to schedule operations in the early stages of structural design, thereby reducing experimental requirements. Thus, an accurate method for forecasting the compressive strength of HPC can significantly reduce time and cost. Many researchers have used mechanics-based simulation methods to quantify the strength of concrete [17], [18], [19], [20], [21], [22]. Rabczuk et al. modeled the fracture of several reinforced concrete structures by using a three-dimensional mesh-free method [18]. Rabczuk and Belytschko applied particle methods to solve several fracture problems involving reinforced concrete structures and the computational results showed good agreement with experimental data [20]. Rabczuk et al. proposed a two-dimensional approach to model the fracture of reinforced concrete structures and took into account the interaction between the concrete and the reinforcement [22]. Drzymałaa used a testing method to investigate the effects of high temperatures on the properties of HPC [23]. Zhao et al. performed an experimental study on the shrinkage of HPC containing fly ash and ground granulated blast-furnace slag [24]. In addition, several linear and nonlinear methods were carried out to find the relationship between the key factors, that may influence the compressive strength of HPC such as cement, fly ash, water, superplasticizer and age of testing [16], [25].

However, these methods make it difficult to obtain an accurate regression function because the compressive strength of HPC is affected by many factors. Also, the properties of concrete have a highly nonlinear relationship with its constituents, which poses difficulties in calculating the compressive strength of HPC from available data [26]. As a result, the common methods used for conventional concrete are often unsatisfactory for forecasting the compressive strength of HPC.

Many Artificial intelligence (AI) techniques have been proposed to solve the aforementioned problem. Chou and Pham introduced ensemble models to forecast the compressive strength of HPC [27]. This ensemble model was created by combining many individual AI techniques. Prasad et al. used an artificial neural network (ANN) model for predicting the compressive strength of self-compacting concrete and HPC [28]. Naderpour et al. predicted the compressive strength of recycled aggregate concrete by using ANN [29]. Ali et al. predicted the compressive strength of ordinary concrete and HPC by using the M5P model tree algorithm [30]. These AI techniques disregard any physical interaction between the input and output variables. In addition, the input parameters of the predictive data should be within the range of input parameters of the trained data, which is a shortcoming of these AI models [31]. These are all promising approaches but they are highly dependent on the initial parameters, which is a strong constraint that inhibits their performance.

Therefore, these AI techniques need to be combined with optimization algorithms and hybrid models [32]. Some authors have proposed these models to solve issues in many fields or areas. Nazari and Sanjayan optimized the parameters of a support vector machine to estimate the geopolymer, mortar and concrete compressive strengths [33]. In their research, five meta heuristic algorithms including the ant colony optimization algorithm, genetic algorithm, imperialist competitive algorithm, artificial bee colony optimization algorithm and particle swarm optimization algorithm, were used to optimize the parameters of the support vector machine (SVM). In another study, Marek applied Bayesian inference to a neural network for forecasting the compressive strength of HPC [34].

Among many optimization algorithms, the firefly algorithm is an efficient optimization tool, which was used to optimize machine learning models in many areas of research. Chou et al. used firefly algorithm-based least square support vector regression to solve many civil engineering prediction problems [35]. Ibrahim and Khatib optimized the random forests technique using the firefly algorithm and applied this model to forecast hourly global solar radiation [36]. However, using the firefly algorithm for enhancing the capability of artificial neural networks (ANN) has not received much attention, especially in civil engineering.

Therefore, this research seeks to apply the modified firefly algorithm (MFA) to optimize the weights and biases of ANN for effectively predicting the compressive strength of HPC. Specifically, the firefly algorithm has been modified for high dimensional optimization and combined with two smart components such as chaotic map and Lévy flights. Moreover, the parameters of ANN are updated, memorized and optimized by MFA during the training process, so the computing time is remarkably reduced. This study also aims to validate the expert system by employing the k-fold cross-validation algorithm. Meanwhile, the performance of MFA-ANN will be compared with that of other techniques employed in similar work by hypothesis testing.

The remaining structure of the paper is divided into five sections. The next section presents a literature review on the current research related to prediction of the compressive and tensile strength of HPC by using machine learning technique. Section 3 describes the research methodology and performance evaluation methods. Section 4 outlines the properties that affected the compressive and tensile strength of high-performance concrete and two experimental datasets used in this study. Section 5 subsequently presents data preprocessing, model application, prediction of results of the MFA-ANN, and compares performance of the model with other methods based on the analytical results. The final section will summarize the research and provide concluding remarks.

Section snippets

Literature review

Forecasting the mechanical properties of concrete such as compressive strength is an important task in civil engineering because it requires many input parameters from various design practices [37], [38]. An efficient and reliable model for estimating the compressive strength in the early stages of a project can certainly shorten project duration [37]. In recent years, many studies using various approaches for estimating the compressive strength of concrete have been reported [39], [40], [41],

Artificial neuron network

ANN simulates the function of the biological neuron by imitating the working principles of the human brain. ANN is based on a set of connected units called artificial neurons as illustrated in Fig. 1. Each neuron transmits a signal to another neuron by a connection or synapse. Each connection is assigned a weight, which can modify the strength of the signal sent downstream [51], [52]. The construction of ANN can be divided into three main steps: (1) defining inputs and outputs of the problem;

Dataset 1: compressive strength of high-performance concrete

The efficiency of the proposed expert system is evaluated using published datasets [26], [63], [64], [65], [66], [67]. Database 1 includes a total of 1133 samples of high-performance concrete with one output variable and eight quantitative input variables. Eight inputs are investigated including the amount of cement, water, blast furnace slag, coarse aggregate, fine aggregate, super plasticizers, fly ash and the age of testing, while the compressive strength (in MPa) is the output.

Each input

Data preprocessing and model application

The results of the performance tests used to predict the compressive and tensile strength of high strength concrete, which were obtained from the MFA-ANN expert system, will be discussed in this section. The K-fold cross-validation method is used in this research to reduce the over-fitting problem in model selection [73]. Kohavi demonstrated that 10-fold is the optimal number of folds that can obtain a good result within an acceptable timeframe [74]. There are several cross-validation methods

Conclusions

We have for the first time investigated an efficient approach based on MFA-ANN for predicting the compressive and tensile strength of high-performance concrete. Two datasets of high-performance concrete samples from various laboratories are used to investigate the efficiency of the proposed expert system. The number of instances of the two datasets are 1133 and 714 samples, respectively. A 10-fold cross-validation method is used to reduce the overfitting problem of system performance. The

Conflict of Interest

None.

Acknowledgements

The first author would like to thank the University of Melbourne for offering the Melbourne Research Scholarship. This work was also supported by The ARC Training Centre for Advanced Manufacturing of Prefabricated Housing (CAMP.H) at the University of Melbourne.

References (78)

  • T. Rabczuk et al.

    A numerical model for reinforced concrete structures

    Int. J. Solids Struct.

    (2005)
  • T. Drzymała

    Effects of high temperature on the properties of high performance concrete (HPC)

    Procedia Eng.

    (2017)
  • Y. Zhao et al.

    Experimental study on shrinkage of HPC containing fly ash and ground granulated blast-furnace slag

    Constr. Build. Mater.

    (2017)
  • H.I. Erdal et al.

    High performance concrete compressive strength forecasting using ensemble models based on discrete wavelet transform

    Eng. Appl. Artif. Intell.

    (2013)
  • I.C. Yeh

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

    Cem. Concr. Res.

    (1998)
  • J.-S. Chou et al.

    Enhanced artificial intelligence for ensemble approach to predicting high performance concrete compressive strength

    Constr. Build. Mater.

    (2013)
  • B.K.R. Prasad et al.

    Prediction of compressive strength of SCC and HPC with high volume fly ash using ANN

    Constr. Build. Mater.

    (2009)
  • H. Naderpour et al.

    Compressive strength prediction of environmentally friendly concrete using artificial neural networks

    J. Build. Eng.

    (2018)
  • A. Behnood

    Prediction of the compressive strength of normal and high-performance concretes using M5P model tree algorithm

    Constr. Build. Mater.

    (2017)
  • K.M. Hamdia

    Predicting the fracture toughness of PNCs: a stochastic approach based on ANN and ANFIS

    Comput. Mater. Sci.

    (2015)
  • A. Nazari et al.

    Modelling of compressive strength of geopolymer paste, mortar and concrete by optimized support vector machine

    Ceram. Int.

    (2015)
  • M. Słoński

    A comparison of model selection methods for compressive strength prediction of high-performance concrete using neural networks

    Comput. Struct.

    (2010)
  • I.A. Ibrahim et al.

    A novel hybrid model for hourly global solar radiation prediction using random forests technique and firefly algorithm

    Energy Convers. Manage.

    (2017)
  • A. Behnood et al.

    Predicting modulus elasticity of recycled aggregate concrete using M5’ model tree algorithm

    Constr. Build. Mater.

    (2015)
  • Z. Yuan et al.

    Prediction of concrete compressive strength: Research on hybrid models genetic based algorithms and ANFIS

    Adv. Eng. Softw.

    (2014)
  • R.J. Thomas et al.

    Stepwise regression modeling for compressive strength of alkali-activated concrete

    Constr. Build. Mater.

    (2017)
  • G. Dogan et al.

    Concrete compressive strength detection using image processing based new test method

    Measurement

    (2017)
  • R. Rebouh

    A practical hybrid NNGA system for predicting the compressive strength of concrete containing natural pozzolan using an evolutionary structure

    Constr. Build. Mater.

    (2017)
  • J. Sobhani

    Prediction of the compressive strength of no-slump concrete: A comparative study of regression, neural network and ANFIS models

    Constr. Build. Mater.

    (2010)
  • H.I. Erdal

    Two-level and hybrid ensembles of decision trees for high performance concrete compressive strength prediction

    Eng. Appl. Artif. Intell.

    (2013)
  • P. Yuvaraj

    Support vector regression based models to predict fracture characteristics of high strength and ultra high strength concrete beams

    Eng. Fract. Mech.

    (2013)
  • A.H. Alavi et al.

    Prediction of principal ground-motion parameters using a hybrid method coupling artificial neural networks and simulated annealing

    Comput. Struct.

    (2011)
  • Q. Liu

    Ensemble artificial neural networks applied to predict the key risk factors of hip bone fracture for elders

    Biomed. Signal Process. Control

    (2015)
  • T.B. Moghaddam

    The use of SVM-FFA in estimating fatigue life of polyethylene terephthalate modified asphalt mixtures

    Measurement

    (2016)
  • T.N. Singh et al.

    Prediction of thermal conductivity of rock through physico-mechanical properties

    Build. Environ.

    (2007)
  • S. Lee

    Background Information of Deep Learning for Structural Engineering

    Arch. Comput. Methods Eng.

    (2017)
  • X.-S. Yang
  • A.H. Gandomi

    Firefly algorithm with chaos

    Commun. Nonlinear Sci. Numer. Simul.

    (2013)
  • W.-C. Hong

    SVR with hybrid chaotic genetic algorithms for tourism demand forecasting

    Appl. Soft Comput.

    (2011)
  • Cited by (0)

    View full text