Sensitivity estimation of first excursion probabilities of linear structures subject to stochastic Gaussian loading

https://doi.org/10.1016/j.compstruc.2021.106482Get rights and content

Highlights

Abstract

This contribution focuses on evaluating the sensitivity associated with first excursion probabilities of linear structural systems subject to stochastic Gaussian loading. The sensitivity measure considered is the partial derivative of the probability with respect to parameters that affect the structural response, such as dimensions of structural elements. The actual calculation of the sensitivity demands solving high dimensional integrals over hypersurfaces, which can be challenging from a numerical viewpoint. Hence, sensitivity evaluation is cast within the context of a reliability analysis that is conducted with Directional Importance Sampling. In this way, the sought sensitivity is obtained as a byproduct of the calculation of the failure probability, where the post-processing step demands performing a sensitivity analysis of the unit impulse response functions of the structure. Thus, the sensitivity is calculated using sampling by means of an estimator, whose precision can be quantified in terms of its standard deviation. Numerical examples involving both small- and large-scale structural models illustrate the procedure for probability sensitivity estimation.

Introduction

A widespread means for quantifying the level of safety of a structure subject to dynamic uncertain loading is the so-called first excursion probability, which measures the chances that one or more responses of interest exceed a prescribed threshold level within the duration of the stochastic excitation [1]. Clearly, the properties of the structure such as dimensions of structural members, stiffnesses or masses affect this probability and hence, it is of paramount interest to calculate the associated sensitivity. That is, to quantify how this probability changes due to modifications of those properties, see e.g. [2], [3], [4], [5], [6]. Such information is of utmost importance, for example, within the context of decision making and risk optimization [7], [8], as it allows determining optimal design solutions which balance construction, operation and eventual failure consequences of engineering systems. In particular, the gradient of the failure probability provides a practical sensitivity measure. Nonetheless, the calculation of both first excursion probability and its sensitivity in terms of a gradient is a challenging task, as the estimation of the probability itself usually demands solving a high dimensional integral that does not possess a closed-form solution.

The estimation of the gradient of a failure probability is a topic which has been addressed in several contributions considering different approaches, see e.g. [9], [10], [11], [12], [13]. However, the number of contributions addressing the specific issue of estimating the gradient of first excursion probabilities of dynamical systems subject to stochastic loading is more restricted and can be classified into two major groups. In the first of these groups, the gradient of the failure probability is calculated with respect to distribution parameters of some structural properties characterized as random variables. In such case, by means of the so-called score function [10], [14], the probability sensitivity of complex nonlinear structural systems has been estimated by means of Subset Simulation, as reported in [15], [16]. The second group of approaches focuses on estimating the gradient of the first excursion probability with respect to deterministic parameters of a structural system. Within this second group, one class of approaches employs stochastic simulation in conjunction with Bayes’ theorem in order to calculate the sought sensitivity measure [17], [18]. Another class of approaches employs stochastic simulation as well, but in combination with local approximations of the responses of interest [19], [20], [21]. Yet another class of approaches considers some approximations of second-order moments [22].

This work focuses on estimating the gradient of the first excursion probability with respect to deterministic parameters (such as dimensions of structural members, stiffnesses or masses) of linear structures subjected to dynamic Gaussian loading. These last considerations are focused on applications that involve mainly structural systems where serviceability conditions are of interest, see e.g. [23], [24], [25]. The theoretical framework for assessing such gradient is based on Directional Sampling [26], [27], [28]. The reason for choosing Directional Sampling over other techniques for reliability analysis lies in the fact that it allows identifying explicitly the limit state hypersurface that separates safe and failure domains. This is most useful for reliability sensitivity estimation [13]. The novel contribution of this work comes into the integration and application of Directional Sampling for performing sensitivity analysis for the specific case of first excursion probabilities of linear structures. Specifically, an Importance Sampling density function developed in [29], [30] is injected within the framework of Directional Sampling, allowing the development of a Directional Importance Sampling (DIS) scheme for probability and probability sensitivity estimation. It is shown that the sought sensitivity becomes a byproduct of the failure probability estimation carried out by means of DIS and that its practical implementation demands performing a sensitivity analysis of the unit impulse response functions of the structure under consideration.

The remaining part of this contribution is organized as follows. Section 2 presents the problem considered in this contribution, namely the calculation of the gradient of a first excursion probability. Then, Section 3 discusses the framework for estimating the aforementioned gradient, which combines concepts of Directional Sampling and Directional Importance Sampling. The application of that framework is illustrated in Section 4, including both small- and large-scale structural models. Finally, Section 5 closes with some conclusions and outlook.

Section snippets

Preliminary remarks

This section states the class of problems considered in this work. In particular, Section 2.2 describes the type of stochastic loading considered and the evaluation of the structural response. Then, Section 2.3 presents formal definitions of both the first excursion probability and its gradient.

Gaussian loading and structural dynamic response

Consider a dynamic load p acting over a structural system, which is represented as a discrete Gaussian process at time instants tk=(k-1)Δt,k=1,,nT, where Δt is the time discretization, nT the number of

General remarks

Approaches for estimating the gradient of the failure probability are usually cast as a post-processing step of a reliability analysis, see e.g. [10], [13], [15], [40], [41]. This makes sense from a practical viewpoint, as one usually requires the information on probability and its sensitivity simultaneously. Furthermore, the information provided by structural analyses performed at the reliability analysis stage can be considered as well for sensitivity analysis. Therefore, this work follows

Overview of the examples

This Section presents two examples that illustrate the application of the framework for the estimation of the probability and its gradient as discussed above. The first example comprises a test model while the second one involves the finite element model of a structural system. In both examples, comparisons with reference results are included, in order to provide a quantitative assessment of the advantages of the framework.

Example 1: Two-degree-of-freedom oscillator subject to discrete white noise excitation

The first example involves a test problem which consists of the

Conclusions and outlook

This contribution has explored the application of Directional Importance Sampling for the estimation of the sensitivity of the first excursion probability of a linear structure subject to stochastic Gaussian loading. The sensitivity measure considered herein is the partial derivative of the failure probability with respect to parameters that affect the structural response. The results obtained in this contribution suggest that:

  • It is possible to estimate the sought derivatives in a

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.

Acknowledgments

This research is partially supported by ANID— (National Agency for Research and Development, Chile) under its program FONDECYT, Grant Nos. 1180271 and 1200087, and Universidad Tecnica Federico Santa Maria – Dirección de Postgrado y Programas under its programs PAC (Programa Asistente Científico 2017) and PIIC (Programa de Incentivo a la Iniciación Científica). This support is gratefully acknowledged by the authors.

References (54)

  • J. Ching et al.

    Local estimation of failure probability function and its confidence interval with maximum entropy principle

    Probab Eng Mech

    (2007)
  • H. Jensen et al.

    Reliability-based optimization of stochastic systems using line search

    Comput Methods Appl Mech Eng

    (2009)
  • M. Valdebenito et al.

    Efficient strategies for reliability-based optimization involving non linear, dynamical structures

    Comput Struct

    (2011)
  • M. Valdebenito et al.

    Reliability sensitivity estimation of linear systems under stochastic excitation

    Comput Struct

    (2012)
  • S. Chakraborty et al.

    Reliability based optimum design of tuned mass damper in seismic vibration control of structures with bounded uncertain parameters

    Probab Eng Mech

    (2011)
  • D. García et al.

    Dynamic behaviour of a timber footbridge with uncertain material properties under a single deterministic walking load

    Struct Saf

    (2019)
  • D. Honfi et al.

    Reliability of beams according to eurocodes in serviceability limit state

    Eng Struct

    (2012)
  • M. Huang et al.

    Optimal performance-based design of wind sensitive tall buildings considering uncertainties

    Comput Struct

    (2012)
  • J. Nie et al.

    Directional methods for structural reliability analysis

    Struct Saf

    (2000)
  • O. Ditlevsen et al.

    Directional simulation in Gaussian processes

    Probab Eng Mech

    (1988)
  • M. Misraji et al.

    Application of directional importance sampling for estimation of first excursion probabilities of linear structural systems subject to stochastic Gaussian loading

    Mech Syst Signal Process

    (2020)
  • G. Stefanou

    The stochastic finite element method: Past, present and future

    Comput Methods Appl Mech Eng

    (2009)
  • G. Schuëller et al.

    A critical appraisal of reliability estimation procedures for high dimensions

    Probab Eng Mech

    (2004)
  • S. Au et al.

    First excursion probabilities for linear systems by very efficient importance sampling

    Probab Eng Mech

    (2001)
  • Z. Lu et al.

    Reliability sensitivity method by line sampling

    Struct Saf

    (2008)
  • M. Valdebenito et al.

    Probability sensitivity estimation of linear stochastic finite element models applying line sampling

    Struct Saf

    (2019)
  • P. Koutsourelakis et al.

    Reliability of structures in high dimensions, part I: Algorithms and applications

    Probab Eng Mech

    (2004)
  • Cited by (5)

    • On the use of Directional Importance Sampling for reliability-based design and optimum design sensitivity of linear stochastic structures

      2022, Probabilistic Engineering Mechanics
      Citation Excerpt :

      In this work, DIS [32] is adopted to evaluate the reliability constraints. Further, this method also provides estimates of the first-order derivatives of the failure probability by reusing the sampling results [34]. As a result, sensitivities with respect to design variables and general model parameters can be obtained as a post-processing step of reliability assessment.

    • Time-Dependent Reliability of Aging Structures: Overview of Assessment Methods

      2021, ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering
    View full text