Finite Element Structural Analysis using Imprecise Probabilities
Based on P-Box Representation
Hao Zhang
1
, R. L. Mullen
2
, and R. L. Muhanna
3
1
University of Sydney, Sydney 2006, Australia, haozhang@usyd.edu.au
2
University of South Carolina, Columbia, SC 29208, USA, rlm@cec.sc.edu
3
Georgia Institute of Technology, Atlanta 31407, USA, rafi.muhanna@gtsav.gatech.edu
Abstract: Imprecise probability identifies a number of various mathematical frameworks for making
decisions when precise probabilities (or PDF) are not known. Imprecise probabilities are normally
associated with epistemic sources of uncertainty where the available knowledge is insufficient to construct
precise probabilities. While there is no “unified theory of imprecise probabilities”, most frameworks
describe the uncertainty in terms of bounded possibilities or unknown probabilities between a specified
lower and upper bounds.
In this paper, the authors’ previous developed interval finite element methods are extended to compute p-
box structures of a finite element solution where loading parameters are described by p-box structures. Both
discrete p-box structures and interval p-box Monte Carlo algorithms are presented along with example
problems that illustrate the capabilities of the new methods. The computational efficiency of the p-box
finite element methods is also presented.
Keywords: Uncertainty; Imprecise Probability; P-Box; Interval Finite Elements.
1. Introduction
Engineering analysis and design require users to make several assumptions, typically at different levels.
One of these levels is the choice of the underlying mathematical model of the system. Another level is the
description of the model parameters. Assumptions are usually made to facilitate processing the analysis and
design, which result in that the nature of engineering computation is conditioned by a priori assumptions. In
a deterministic analysis, the geometry, loads, and material properties are assumed to have specific values.
Conversely, if these assumptions do not hold, a new model with a new set of assumptions has to be used to
reflect the real variations in geometry, load and material properties. The best candidate methods for
handling this new situation are probabilistic methods. However, they assume complete information about
the variability of all parameters and their respective Probability Density Functions (PDFs) are available.
The available information, however, might range from scarce or limited to comprehensive. When there is
limited or insufficient data, designers fall back to deterministic analysis, which is an irrational approach that
does not account for parameters variability. On the other hand when more data is available but insufficient
to completely justify a particular PDF, designers would assume a probability distribution that might not
represent the real behavior of the system. More advanced Bayesian methods can treat the incomplete
information in describing PDFs by updating parameters of a probability distribution. With a Bayesian
approach, subjective judgments are required to estimate the random variables (i.e., form of the PDF), and
thus the approach remains a subjective representation of uncertainty. In such a case, we find ourselves in
4th International Workshop on Reliable Engineering Computing (REC 2010)
Edited by Michael Beer, Rafi L. Muhanna and Robert L. Mullen
Copyright © 2010 Professional Activities Centre, National University of Singapore.
ISBN: 978-981-08-5118-7. Published by Research Publishing Services.
doi:10.3850/978-981-08-5118-7 013
211