Structural reliability analysis using Monte Carlo simulation and neural networks

Offshore riser systems are subjected to wind, wave and current loadings, which are random in nature. Nevertheless, the current deterministic based design and analysis practice could not quantitatively evaluate the safety of structures taking random environmental loadings into consideration, due to high computational costs. Structural reliability method, as an analysis tool to quantify probability of failure of components or systems, can account for uncertainties in environmental conditions and system parameters. It is particularly useful in cases where limited experience exists or a risk-based evaluation of design is required. Monte Carlo Simulation (MCS) method is the most widely accepted method and usually used to benchmark other proposed reliability methods. However, MCS is computationally demanding for predicting low failure probabilities, especially for offshore dynamic problems involving many types of uncertainties. Innovative structural reliability methods are desired to perform reliability analysis, so as to predict the low failure probabilities associated with extreme values. Variety of structural reliability methods are proposed in the literature to reduce the computational burden of MCS. The post processing methods, which recover PDF or tail distribution of random variable from sample data to perform structural reliability analysis, have great advantages over the methods from other categories on solving engineering problems. Thus the main focus of our study is on post processing structural reliability methods. In this paper, four post processing reliability methods are compared on the prediction of low failure probabilities with applications to a drilling riser system and a steel catenary riser (SCR) system: Enhanced Monte Carlo Simulation (EMCS) assumes the failure probability follows the asymptotic behavior and uses high failure probabilities to predict low failure probabilities; Multi-Gaussian Maximum Entropy Method (MGMEM) assumes the probability density function (PDF) is a summation of Gaussian density functions and adopts maximum entropy methods to obtain the model parameters; Shifted Generalized Lognormal Distribution (SGLD) method proposes a distribution that specializes to the normal distribution for zero skewness and is able to assume any finite value of skewness for versatility; and Generalized Extreme-Value Distribution method (GEV) comprises three distribution families: the Gumbel-type, Frechet-type and Weibull-type distribution. The study compares the bias errors (the difference between the predicted values and the exact values) and variance errors (the variability of the predicted values) of these methods on the prediction of low failure probabilities with applications to two riser systems. This study could provide offshore engineers and researchers feasible options for marine riser system structural reliability analysis.

Download Full-text

Shallow and Deep Artificial Neural Networks for Structural Reliability Analysis

ASCE-ASME J Risk and Uncert in Engrg Sys Part B Mech Engrg ◽

10.1115/1.4047636 ◽

2020 ◽

Author(s):

Wellison J. S. Gomes

Keyword(s):

Neural Networks ◽

Artificial Neural Networks ◽

Reliability Analysis ◽

Structural Reliability ◽

Deep Neural Networks ◽

Surrogate Models ◽

The Other ◽

Computational Costs ◽

Structural Reliability Analysis ◽

Artificial Neural

Abstract Surrogate models are efficient tools which have been successfully applied in structural reliability analysis, as an attempt to keep the computational costs acceptable. Among the surrogate models available in the literature, Artificial Neural Networks (ANNs) have been attracting research interest for many years. However, the ANNs used in structural reliability analysis are usually the shallow ones, based on an architecture consisting of neurons organized in three layers, the so-called input, hidden and output layers. On the other hand, with the advent of deep learning, ANNs with one input, one output, and several hidden layers, known as deep neural networks, have been increasingly applied in engineering and other areas. Considering that many recent publications have shown advantages of deep over shallow ANNs, the present paper aims at comparing these types of neural networks in the context of structural reliability. By applying shallow and deep ANNs in the solution of four benchmark structural reliability problems from the literature, employing Monte Carlo simulation and adaptive experimental designs, it is shown that, although good results are obtained for both types of ANNs, deep ANNs usually outperform the shallow ones.

Download Full-text