New Gene Expression Programming models for normalized shear modulus and damping ratio of sands

https://doi.org/10.1016/j.engappai.2015.07.022Get rights and content

Highlights

  • Paper proposes several new models based on GEP for dynamic properties of sands.

  • Models can predict normalized shear modulus and damping ratio of sands.

  • Proposed equations have been measured through published experimental data.

  • Comparison between predicted and measured values showed high accuracy of models.

  • The accuracy of other common models was assessed through experimental data.

Abstract

As the most important dynamic properties of soils, shear modulus and damping ratio are two parameters employed to solve problems including seismic site response evaluation, dynamic analyses and equivalent-linear models. The work presented in this paper proposes two models for evaluation of the normalized shear modulus and two additional models for evaluation of the damping ratio of sands through Gene Expression Programming (GEP). The data used in the modeling entails the valid experimental results obtained from previous researchers. As compared to the secondary models, the first two models are more accurate with larger equation length. The parameters taken into account as model inputs consisted of shear strain, mean effective confining pressure, and void ratio. In order to evaluate the performance and accuracy, the proposed models were processed through several statistical measures such as Mean Square Error (MSE), Root Mean Square Error (RMSE) and coefficient of determination (R2). Furthermore, the relative difference between predicted and measured values was calculated, which suggested that the models were desirably accurate. Finally, the model outputs were compared against other studies, the results of which demonstrated that the proposed models are capable of estimating the dynamic parameters of sands more accurately.

Introduction

The nature and distribution of earthquake hazards are extremely influenced by soil response under cyclic loading. To a large extent, such a response is controlled by the dynamic characteristics of the soil. Dynamic characteristics of geotechnical materials are often represented by shear modulus (G) and damping ratio (D). In order to examine the effects of the dynamic loads on civil engineering systems, it is essential to properly understand soil behavior, attain precision measurements and quantitative models describing the soil dynamic characteristics (Kramer, 1996). Moreover, the description and identification of soil are necessary prior to carrying out any geotechnical design. On the other hand, understanding the dynamic properties of soil allows a geotechnical engineer to more accurately evaluate and monitor the soil parameters. The dynamic response of soils is employed to solve several problems including slope stability, soil-structure interaction, machinery vibrations and seismic stability of structures under sea wave, wind, traffic and other dynamic loads. The relations between soil dynamic parameters are crucial for solving the abovementioned problems. Since there is a wide range of equations associated with the soil damping ratio and shear modulus, the selection of an equation greatly affects the results of engineering analyses. For that reason, the newest and most accurate methods need to be employed so as to achieve minimum error margin.

The experimental evaluation of the shear modulus and damping ratio of soils have so far been measured by numerous researchers through various devices such as Resonant Column (RC) (Hardin and Drnevich, 1972a, Khan et al., 2008, Khan et al., 2011, Senetakis et al., 2012, Wilson, 1988), Cyclic Triaxial (CT) (Khan et al., 2011, Yasuda and Matsumoto, 1993, Yoshimi et al., 1984), Cyclic Simple Shear (CSS) (Lanzo et al., 1997), cyclic simple Torsional Shear (TS) (Yasuda and Matsumoto, 1993) and combined device Resonant Column Torsional Shear (RCTS) (Darendeli, 2001, Lee, 2000, Menq, 2003). The significance of parameters contributing to the dynamic properties of soils has been reported by Hardin and Drnevich (1972b) and Darendeli (2001). The most important parameters contributing to shear modulus include shear strain (γ), mean effective confining pressure (σ¯) and soil conditions (i.e., D50, etc.) as well as parameters contributing to damping ratio, not to mention the number of loading cycles (N) (Darendeli, 2001, Hardin and Drnevich, 1972b, Ishibashi and Zhang, 1993, Iwasaki et al., 1978, Kokusho, 1980, Stokoe et al., 1999).

Iwasaki et al. (1978) and Kokusho (1980) studied the effect of void ratio, effective confining pressure and shear strain percentage on the dynamic parameters of sands. Moreover, Seed et al. (1986) assessed through field and laboratory experiments the impact of mean effective confining pressure, relative density and shear strain percentage on granular soils. Ishibashi and Zhang (1993) collected different laboratory data for conducting dynamic evaluation of shear modulus and damping ratio in non-plastic sands and high-plasticity clays. They offered a series of equations for soil shear modulus and damping ratio as functions of mean effective confining pressure, plasticity index and shear strain. Their equations are valid within confining pressures between 0.2 and 10 atms (Darendeli, 2001). Darendeli (2001) examined the effect of soil type conditions, loading frequency, loading cycles (f and N) and plasticity index (PI) for a broad range of soils including gravel, sand and clay at various plasticity rates. Zhang et al. (2005) introduced several equations for estimating normalized shear modulus and damping ratio of Quaternary, Tertiary and older, and residual/saprolite soils. The parameters used in equations proposed for shear modulus include shear strain range, plasticity index and confining pressure, where the proposed equation for damping ratio comprises a term as a minimum damping ratio added to a polynomial function of normalized shear modulus. In all the mentioned studies, two or three of these parameters (e, σ¯ and γ%) have been introduced as contributing factors to dynamic properties of sands.

Nonlinear optimization methods have been used in several problems. In the remediation of soil contamination (Chen et al., in press), an optimization system for surfactant-enhanced aquifer remediation has been developed by Qin et al. (2007). As a powerful method for nonlinear optimization, Genetic Algorithm has been successfully utilized in civil engineering analyses, particularly geotechnical engineering such as numerical modeling of stress–strain behavior under cyclic loading (Shahnazari et al., 2010), liquefaction-induced displacement (Javadi et al., 2006), dynamic soil properties (Cevik and Cabalar, 2009) and simulation of static soil shear strength (Mousavi et al., 2011). The main objective of this study is to provide new mathematical models based on GEP, which can predict normalized shear modulus (G/Gmax) and damping ratio (D%) of sands more accurately with minimum error margin, as compared to previous models. The parameters e, σ¯ and γ were considered as inputs of the proposed models. Through mathematical models from several researchers including Darendeli (2001), Rollins et al. (1998) and Ishibashi and Zhang (1993) and published experimental data, the validity and performance of these models were compared and demonstrated to be more accurate than previous models. Using the collected data, the accuracy of Darendeli (2001), Rollins et al. (1998) and Ishibashi and Zhang (1993) models was also assessed in the estimation of the normalized shear modulus and damping ratio of sands.

Section snippets

Datasets

The datasets used in this study entailed the valid experimental findings obtained by various researchers which are an outcome of several devices including Resonant Column (RC) (Moayerian, 2012, Saxena and Reddy, 1989, Senetakis et al., 2013), Cyclic Triaxial (CT) (Kokusho, 1980, Kokusho, 2004, Rollins et al., 1998), Cyclic Simple Shear (CSS) (Anderson, 2003, D’Elia et al., 2003, Lanzo et al., 1997), cyclic simple Torsional Shear (TS) (Iwasaki et al., 1978, Uthayakumar, 1992) and combined device

Gene Expression Programming

As an optimization method inspired by the theory of evolution and natural selection, Genetic Algorithm (GA) was recognized as a stochastic optimization technique (Holand, 1975), which has been further developed by Goldberg (1989). Symbolic strings of fixed length (chromosomes) are the unique solutions in Genetic Algorithm; they are assessed through a fitness function, where the algorithm terminates once the output requirement is met. In fact, fitness function is one of the most important

Proposed models

In GEP, the values of setting parameters have important influence on the fitness of the output model. These include the number of genes, number of chromosomes and gene׳s head size, and the rate of genetic operators. Since there was little information available about the genetic parameters, the software was implemented by various adjustments through trial and error. Moreover, the performance of each model was assessed based on statistical measures such as RMSE, MSE and R2. As a result, the most

Accuracy evaluation of the proposed models

The accuracy of the proposed models for training, testing and total data was compared using statistical measures, the results of which can be seen in Table 3, Table 4. These results shed light on the fact that accuracies of the proposed models are almost identical in terms of statistical measures. Hence, both models are sufficiently accurate for estimation of the normalized shear modulus and damping ratio.

In Fig. 2, Fig. 3, a comparison was drawn between the total data for values predicted by

Conclusion

This paper mainly intends to propose several models based on the Gene Expression Programming (GEP), for predicting the normalized strain modulus (G/Gmax) and damping ratio (D%) of sands as functions of mean effective confining pressure, void ratio and shear strain percentage. These models have been measured through valid published experimental data using a number of devices including Resonant Column (RC), Cyclic Triaxial (CT), Cyclic Simple Shear (CSS), cyclic simple Torsional Shear (TS) and

References (41)

  • Y. Yoshimi et al.

    Undrauined cyclic shear strength of a dense Niigata sand

    Soils Found.

    (1984)
  • Anderson, D.G., 2003. Laboratory testing of nonlinear soil properties: I & II, Report, CH2M HILL, Bellevue, Washington...
  • D.L. Chambers

    The Practical Handbook of Genetic Algorithms Applications

    (2001)
  • Chen, M., Xu, P., Zeng, G., Yang, C., Huang, D., Zhang, J. Bioremediation of soils contaminated with polycyclic...
  • D’Elia, B., Lanzo, G., Pagliaroli, A., 2003. Small strain stiffness and damping of soils in a direct simple shear...
  • M.B. Darendeli

    Development of a New Family of Normalized Modulus Reduction and Material Damping Curves

    (2001)
  • Ferreira, C., 2001. Gene expression programming in problem solving. In: Roy, R., Koeppen, M., Ovaska, S., Furuhashi,...
  • Ferreira, C., 2004. Gene expression programming and the evolution of computer programs. In: Castro, L.N.D., Zuben,...
  • C. Ferreira

    Gene Expression Programming: Mathematical Modeling by an Artificial Intelligence

    (2006)
  • D.E. Goldberg

    Genetic Algorithms in Search, Optimization and Machine Learning

    (1989)
  • Cited by (32)

    • Prediction of secant shear modulus and damping ratio for an extremely dilative silica sand based on machine learning techniques

      2023, Soil Dynamics and Earthquake Engineering
      Citation Excerpt :

      Dynamic shear modulus and damping ratio of soils are two important soil dynamic properties used in geotechnical earthquake engineering issues, including soil-structure interaction analysis [1–3] and dynamic bearing capacity of machine foundations [4–6], .

    • Prediction of mode I fracture toughness of rock using linear multiple regression and gene expression programming

      2022, Journal of Rock Mechanics and Geotechnical Engineering
      Citation Excerpt :

      As a result, the remaining chromosomes are changed into new chromosomes. This process continues until the stopping conditions (maximum number of generations or achieving the suitable solution) are reached (Ferreira, 2001; Keshavarz and Mehramiri, 2015; Faradonbeh et al., 2016a; Khandelwal et al., 2016, 2017; Monjezi et al., 2016; Armaghani et al., 2018). In this paper, a total of 60 datasets (see Table 1) were collected from various published studies.

    View all citing articles on Scopus
    View full text