Elsevier

Automation in Construction

Volume 36, December 2013, Pages 136-144
Automation in Construction

Numerical modeling of concrete strength under multiaxial confinement pressures using linear genetic programming

https://doi.org/10.1016/j.autcon.2013.08.016Get rights and content

Highlights

  • LGP is introduced for modeling of concrete under multiaxial compression.

  • The models are externally validated using several statistical criteria.

  • The proposed models outperform several models found in the literature.

Abstract

New numerical models are developed to predict the strength of concrete under multiaxial compression using linear genetic programming (LGP). The models are established based on a comprehensive database obtained from the literature. To verify the applicability of the derived models, they are employed to estimate the strength of parts of the test results that are not included in the modeling process. The external validation of the model is further verified using several statistical criteria. The results obtained by the proposed models are much better than those provided by several models found in the literature. The LGP-based equations are remarkably straightforward and useful for pre-design applications.

Introduction

High strength concrete, high performance concrete and ultra-high performance concrete are nowadays widely used in civil engineering tasks. This implies the necessity of having a better understanding of the relationship between the mechanical properties of these materials. One of the major technical interests for different concrete types is the failure mechanisms under uniaxial and triaxial loading. On the other hand, as known, there is a wide variety of investigations which have focused on the development and strengthening of structural elements under confinement pressures. Steel fibers, fiber-reinforced polymer (FRP) composites and many other types of composites are the devices which improve the loading capabilities of structural concrete elements under confinement pressures [1], [2], [3], [4]. However, according to literature, there is a huge gap between the experimental and the numerical knowledge of failure mechanisms for the triaxially compressed concrete and composite structures. Hence, it is necessary to develop some equations correlating the confining pressures, uniaxial compressive strength, peak triaxial strengths and other mechanical characteristics for an efficient design of the triaxially compressed concrete structures. In this context, Cordon and Gillespie [5] performed triaxial tests on the plain concrete under multiaxial loading conditions to investigate the effect of different parameters on the strength of concrete. Lahlou et al. [6] presented the preliminary results of an experimental investigation on the confined high strength concrete. In this manner, Chern et al. [7] studied the response of fiber-reinforced concrete under multiaxial loading conditions. Xie et al. [8] performed a series of triaxial tests on the high strength concrete specimens. Also, Attard and Setunge [9] and Setunge et al. [10] performed a series of triaxial tests on larger cylindrical samples involving concretes with strengths as high as 109 MPa. Bohwan et al. [11] presented the preliminary results of an experimental investigation on confined concrete. Imran and Pantazopoulou [12] studied the behavior of plain concrete under triaxial stress states and the damage characteristics of concrete under dry and saturated conditions. In a similar manner, Nielsen [13] prepared an experimental study on the assessment of the triaxial behavior of ultra-high strength mortar. Some researchers performed a series of triaxial tests on the high strength concrete specimens involving concretes with strengths as high as 103 MPa [14], [15]. Candappa et al. [16], [17] performed a series of triaxial tests under low confining pressures on the four high-strength concretes with uniaxial compressive strengths of 41.9 to 103.3 MPa. Moreover, Mei et al. [18] performed an experimental investigation on the stress–strain characteristics of steel sleeve confined high strength concrete. Sfer et al. [19] conducted an experimental study of the confined compression behavior of concrete subjected to hydrostatic pressure for triaxial tests. The behavior of high strength concrete and steel fiber reinforced high strength concrete (1% steel fiber volume fraction) under triaxial compression was investigated by other researchers [20], [21], [22]. Recently, Farnam et al. [23] assessed multiaxial compressive behaviors of high strength concrete, high performance fiber reinforced concrete (HPFRC) and slurry infiltrated fiber concrete (SIFCON) under different confining pressures.

Prediction of the properties of concrete using statistical empirical relationship and numerical methods is a difficult task due to the extreme complexity involved [24], [25], [26], [27]. Besides, the current in-use empirical and numerical models are based on the theories of strength of material that in some cases cannot release the exact and precise results. This is caused due to the lack of data and insufficient knowledge of the micro and macro properties of materials. Such models are derived on the basis of several assumptions, approximations, and simplifications. An efficient idea to cope with this problem is to use machine learning methods. These systems learn adaptively from experience and extract various discriminators. They rely on the real data and natural rules in the nature. In most cases, they can release more exact, precise and reliable results than conventional modeling approaches [26]. Thus, they can be used as valuable tools for facilitating the computer-aided design of various civil engineering projects. Artificial neural networks (ANNs) are the most widely used machine learning methods. ANNs and other machine learning methods have been applied to assess different concrete properties [24], [25], [28], [29], [30]. Despite the acceptable performance of ANNs, they usually do not give a certain function to calculate the outcome using the input values.

Genetic programming (GP) [31] is a new machine learning approach for behavioral modeling of structural engineering problems. The main advantage of the GP-based methods is their ability to generate prediction equations without assuming prior form of the existing relationship. Recently, a robust variant of GP, namely gene expression programming was utilized to predict the ultimate strength of concrete under triaxial compression loading [32]. The triaxial strength of concrete was formulated in terms of mix design parameters such as triaxial confining pressure, ratio of water and superplasticizer summation to binder, ratio of coarse aggregate content to fine aggregate content, fiber index, and age of specimens [32]. Linear genetic programming (LGP) [33], [34] is a new subset of GP with a linear structure similar to the DNA molecule in biological genomes. Unlike classic GP and other soft computing tools like neural networks, the LGP applications to solve problems in civil engineering are restricted to fewer areas [35], [36].

The machine learning methods use the construction information database to automate and computerize the design stages without any need to perform the experimental and numerical tests. The main contribution of this paper to the domain knowledge is to propose and validate the novel machine learning technique, LGP, to automate concrete design for civil infrastructure and building construction. More specifically, the LGP technique is utilized to build predictive models for the strength of concrete under triaxial compression. The concrete strength is formulated in terms of uniaxial compressive strength and confining pressure. The predictions made by models are further compared with those provided by several models found in the literature [8], [9], [10], [11], [14], [15], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], [49], [50].

Section snippets

Genetic programming

GP is a developing subarea of evolutionary algorithms inspired from Darwin's evolution theory. It may generally be defined as a supervised machine learning technique that searches a program space instead of a data space [31], [51]. GP is an extension of genetic algorithms (GAs). Most of the genetic operators used in GA can also be implemented in GP with minor changes. The main difference between GP and GA is the representation of the solution. The GP solutions are computer programs that are

Failure models

Table 1 presents the empirical strength criteria proposed for high strength concretes obtained from the literature [32].

Experimental database

The model is developed based on a comprehensive database obtained from the literature. The gathered database contains 370 test results for the strength capacity of concrete under triaxial compression [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23]. Table 2 shows the geometrical and mechanical properties of the samples

Model validity

Smith [61] states that if a model gives R > 0.8, there is a strong correlation between the predicted and measured values. The model can therefore be judged as very good. Based on the results, the proposed LGP models with low RMSE and MAE, and high R values are able to predict the target values to an acceptable degree of accuracy. Frank and Todeschini [62] argue that the minimum ratio of the number of objects over the number of selected variables for model acceptability is 3. They also suggest

Conclusion

A robust variant of GP, namely LGP, was utilized to formulate the strength capacity of the concrete under confinement conditions. Accurate empirical models were derived for the prediction of the concrete ultimate triaxial strength. A reliable database from the previously published triaxial test results was used for developing the models. The proposed LGP-based models are capable of predicting the ultimate strength of concrete under triaxial compression with high accuracy. The validity of the

References (68)

  • J.S. Chou et al.

    Concrete compressive strength analysis using a combined classification and regression technique

    Autom. Constr.

    (2012)
  • M.A. Kewalramani et al.

    Concrete compressive strength prediction using ultrasonic pulse velocity through artificial neural networks

    Autom. Constr.

    (2006)
  • A. Ilc et al.

    New numerical procedure for the prediction of temperature development in early age concrete structures

    Autom. Constr.

    (2009)
  • M.Y. Cheng et al.

    High performance concrete compressive strength prediction using time-weighted evolutionary fuzzy support vector machines inference model

    Autom. Constr.

    (2012)
  • H.Md. Azamathulla et al.

    Linear genetic programming to scour below submerged pipeline

    Ocean Eng.

    (2011)
  • A.H. Alavi et al.

    Formulation of flow number of asphalt mixes using a hybrid computational method

    Constr. Build. Mater.

    (2011)
  • A.H. Alavi et al.

    Discussion on soft computing approach for real-time estimation of missing wave heights

    Ocean Eng.

    (2010)
  • F.D. Francone et al.

    Extending the boundaries of design optimization by integrating fast optimization techniques with machine-code-based, linear genetic programming

    Inform. Sci.

    (2004)
  • A. Rashed et al.

    Nonlinear modeling of soil deformation modulus through LGP-based interpretation of pressuremeter test results

    Eng. Appl. Artif. Intell.

    (2012)
  • A. Golbraikh et al.

    Beware of q2

    J. Mol. Graphics Modell.

    (2002)
  • A. Peyvandi et al.

    Surface-modified graphite nanomaterials for improved reinforcement efficiency in cementitious paste

    Carbon

    (2013)
  • A. Peyvandi et al.

    Effect of the cementitious paste density on the performance efficiency of carbon nanofiber in concrete nanocomposite

    Constr. Build. Mater.

    (2013)
  • A. Peyvandi et al.

    Enhancement of the structural efficiency and performance of concrete pipes through fiber reinforcement

    Constr Build Mater

    (2013)
  • A. Peyvandi et al.

    Enhancement of the durability characteristics of concrete nanocomposite pipes with modified graphite nanoplatelets

    Constr Build Mater

    (2013)
  • W.A. Cordon et al.

    Variables in concrete aggregates and Portland cement paste which influence the strength of concrete

    ACI Struct. J.

    (1963)
  • J.C. Chern et al.

    Behavior of steel fiber reinforced concrete in multiaxial loading

    ACI Mater. J.

    (1992)
  • M.M. Attard et al.

    Stress–strain relationship of confined and unconfined concrete

    ACI Mater. J.

    (1996)
  • S. Setunge et al.

    Ultimate strength of confined very high-strength concretes

    ACI Struct. J.

    (1993)
  • O. Bohwan et al.

    Experimental study of 60 MPa concrete under triaxial stress

  • I. Imran et al.

    Experimental study of plain concrete under triaxial stress

    ACI Mater. J.

    (1996)
  • C.V. Nielsen

    Triaxial behavior of high-strength concrete and mortar

    ACI Mater. J.

    (1998)
  • F. Ansari et al.

    High-strength concrete subjected to triaxial compression

    ACI Mater. J.

    (1998)
  • Q. Li et al.

    High-strength concrete in triaxial compression by different sizes of specimens

    ACI Mater. J.

    (2000)
  • D.C. Candappa et al.

    Complete triaxial stress strain curves of high-strength concrete

    J. Mater. Civ. Eng.

    (2001)
  • Cited by (35)

    • An explainable machine learning model to predict and elucidate the compressive behavior of high-performance concrete

      2021, Results in Engineering
      Citation Excerpt :

      On a construction project, early prediction of concrete strength is essential for the planning and scheduling of critical construction tasks such as the casting of concrete slab, and removal of concrete form [4]. Developing robust prediction models for HPCCS will provide vital data needed by structural engineers to design quality and successful concrete mixes [1] and proportion of supplementary materials [13] that can bring about remarkable savings in time and cost [25,29]. Thus, accurate and early prediction of concrete strength is critical in concrete construction [4].

    • Prediction of seismic damage spectra using computational intelligence methods

      2021, Computers and Structures
      Citation Excerpt :

      Unlike ANNs and classic regression models, GP is capable of formulating the relationship regardless of a predefined structure. GP and its variants have been successfully used to solve numerous engineering problems (e.g. [23,25,47]. Multi-gene genetic programming (MGGP), one of the variants of GP, was proposed by Searson et al. [48] to improve the performance of the standard GP, where the capability of conventional regression is added.

    View all citing articles on Scopus
    View full text