Elsevier

Computers and Geotechnics

Volume 76, June 2016, Pages 33-42
Computers and Geotechnics

Research Paper
An efficient quasi-Newton approximation-based SORM to estimate the reliability of geotechnical problems

https://doi.org/10.1016/j.compgeo.2016.02.003Get rights and content

Abstract

The first order reliability method (FORM) is efficient, but it has limited accuracy; the second order reliability method (SORM) provides greater accuracy, but with additional computational effort. In this study, a new method which integrates two quasi-Newton approximation algorithms is proposed to efficiently estimate the second order reliability of geotechnical problems with reasonable accuracy. In particular, the Hasofer–Lind–Rackwitz–Fiessler–Broyden–Fletcher–Goldfarb–Shanno (HLRF–BFGS) algorithm is applied to identify the design point on the limit state function (LSF), and consequently to compute the first order reliability index; whereas the Symmetric Rank-one (SR1) algorithm is nested within the HLRF–BFGS algorithm to compute good approximations, yet with a reduced computational effort, of the Hessian matrix required to compute second order reliabilities. Three typical geotechnical problems are employed to demonstrate the ability of the suggested procedure, and advantages of the proposed approach with respect to conventional alternatives are discussed. Results show that the proposed method is able to achieve the accuracy of conventional SORM, but with a reduced computational cost that is equal to the computational cost of HLRF–BFGS-based FORM.

Introduction

Reliability analyses have been proposed as a rational complement to deterministic geotechnical design [1], as they can more directly quantify the influence of the uncertainty about input parameters and their correlation relationships (see e.g., [2], [3]).

Due to its simplicity and efficiency, the first order reliability method (FORM)—that linearly approximates the limit state function (LSF) to estimate the probability of failure—has been widely used in geotechnical reliability analyses (see e.g., [4], [5], [6], [7]). However, the linearization that is inherent to FORM introduces errors in many cases (see e.g., [7], [8]), and the second order reliability method (SORM)—which extends FORM to consider the curvatures of the LSF, hence providing a better approximation—has also been employed as an alternative.

Brza̧kała and Puła [9] and Bauer and Puła [10] analyzed the probability of foundation settlements exceeding an allowable threshold using SORM and a polynomial response surface method (RSM)-based SORM, respectively; Cho [11] combined an artificial neural network (ANN)-based RSM and SORM to compute the reliability of slopes; Lü and co-authors [2], [12], [13] employed various RSMs with SORM to analyze tunnel supports; Chan and Low [14] introduced a practical SORM for foundation reliability analysis using a point-fitted paraboloid method; and Zeng and co-authors [8], [15] applied SORM to evaluate the system reliability of tunnels and slopes, respectively. However, these methods are often more computationally expensive than FORM, due to the need to evaluate the curvatures of the LSF or to construct the response surface function. Therefore, an approach that aims to combine the higher accuracy of SORM-based reliability solutions with the lower computational cost of traditional FORM-based approaches is considered as a useful contribution to the geotechnical field.

This paper proposes an attempt in that direction. In particular, our proposed approach uses the recently proposed Hasofer–Lind–Rackwitz–Fiessler–Broyden–Fletcher–Goldfarb–Shanno (HLRF–BGFS) algorithm [16] to locate the FORM design point efficiently; and it integrates such algorithm with the Symmetric Rank-one (SR1) algorithm [17] to approximate the Hessian matrix (i.e., the second order derivative matrix). The goal is that the identified design point can be used, together with the approximated Hessian matrix, to efficiently estimate the second order probability of failure. Details of the algorithms are discussed first, and then the performance of the proposed method is tested using three typical geotechnical example cases taken from the literature.

Section snippets

Conventional SORM

Conventional SORM estimates the second order probability of failure using (i) the design point and gradient information computed by FORM and (ii) the Hessian matrix computed using finite differences. For completeness, traditional methods to compute FORM solutions and the Hessian matrix, as well as to compute the SORM reliability based on them, are discussed below. Additional details can be found in traditional reliability references such as [18] and [19].

A proposed quasi-Newton approximation-based SORM

In our approach, two types of quasi-Newton approximation—the BFGS and SR1 algorithms—are combined for a more efficient estimation of the second order probability of failure. (Quasi-Newton methods are alternatives to “full” Newton methods, which approximate the Hessian matrices needed at every iteration of gradient-based optimization approaches [38].) In particular, the BFGS algorithm is used, together with the original HLRF algorithm proposed by Hasofer and Lind [22] and Rackwitz and Flessler

Case studies

The reliability of three typical geotechnical problems—settlement of a rectangular foundation, bearing capacity of a shallow footing, and stability of a layered soil slope—are considered in this study as benchmark tests of the proposed method, in which the iHLRF algorithm, the HLRF–BFGS algorithm, and conventional SORM are employed for comparison with our proposed method. The same convergence criteria given in Eq. (15), with ε=0.001, are used for these four methods to make results comparable.

Summary and conclusions

A new method has been proposed in this study for an efficient estimation of the second order probability of failure of geotechnical problems. To reduce the computational cost, the method builds on both the HLRF–BGFS and SR1 algorithms; in particular, instead of using the Hessian matrix computed with a forward difference scheme that is traditionally employed in SORM, it uses good approximations to the Hessian matrix provided by the SR1 algorithm, and it incorporates them to the HLRF–BGFS

Conflict of interest

This work does not have any conflict of interest that needs to be reported.

Acknowledgments

This research was partially supported by the Spanish Ministry of Economy and Competitiveness under Grant BIA2015-69152R and the State Key Laboratory of Geohazard Prevention and Geoenvironment Protection (Grant Nos. SKLGP2011Z002 and SKLGP2015Z016). The first author was supported by China Scholarship Council (CSC) and, for insurance coverage, by Fundación José Entrecanales Ibarra. Dr. Solange Regina and Dr. Gislaine Periçaro kindly provided their MATLAB codes for the HLRF–BGFS algorithm. Their

References (42)

  • Q. et al.

    Reliability analysis of ground–support interaction in circular tunnels using the response surface method

    Int J Rock Mech Min

    (2011)
  • P. Zeng et al.

    System reliability analysis of layered soil slopes using fully specified slip surfaces and genetic algorithms

    Eng Geol

    (2015)
  • G.A. Periçaro et al.

    HLRF–BFGS optimization algorithm for structural reliability

    Appl Math Model

    (2015)
  • R. Rackwitz et al.

    Structural reliability under combined random load sequences

    Comput Struct

    (1978)
  • P.L. Liu et al.

    Optimization algorithms for structural reliability

    Struct Saf

    (1991)
  • G.Q. Cai et al.

    Refined second-order reliability analysis

    Struct Saf

    (1994)
  • H.U. Köylüoǧlu et al.

    New approximations for SORM integrals

    Struct Saf

    (1994)
  • A.R. Zolfaghari et al.

    Simple genetic algorithm search for critical non-circular failure surface in slope stability analysis

    Comput Geotech

    (2005)
  • R. Rackwitz

    Reliability analysis—a review and some perspectives

    Struct Saf

    (2001)
  • J. Duncan

    Factors of safety and reliability in geotechnical engineering

    J Geotech Geoenviron Eng

    (2000)
  • P. Zeng et al.

    Reliability analysis of circular tunnel face stability obeying Hoek–Brown failure criterion considering different distribution types and correlation structures

    J Comput Civ Eng

    (2014)
  • Cited by (20)

    • An efficient method combining polynomial-chaos kriging and adaptive radial-based importance sampling for reliability analysis

      2021, Computers and Geotechnics
      Citation Excerpt :

      Currently, methods for reliability analysis are increasingly being developed and can be generally divided into three categories, i.e., approximation methods, simulation methods, and metamodel-based methods (Marelli et al., 2021; Moustapha et al., 2021; Schöbi et al., 2021). Approximation methods, such as the first-order reliability method (FORM) (Hasofer and Lind, 1974; Rackwitz and Flessler (1978); Zhang and Der Kiureghian, 1995) and second-order reliability method (SORM) (Breitung, 1989; Cai and Elishakoff, 1994, Zeng et al. 2016), are based on approximating the original computational model (also termed the limit-state function) by using a linear or quadratic Taylor expansion at a reference point. These methods have been shown to be efficient for simple problems, but they appear to become unstable in cases of a complex nonlinear limit-state function.

    View all citing articles on Scopus
    1

    Department of Ground Engineering, Technical University of Madrid, Madrid, Spain

    View full text