Robust Estimation of Reliability in the Presence of Multiple Failure Modes

In structural design, every component or system needs to be tested to ascertain that it satisfies the desired safety levels. Due to the uncertainties associated with the operating conditions, design parameters, and material systems, this task becomes complex and expensive. Typically these uncertainties are defined using random, interval or fuzzy variables, depending on the information available. Analyzing components or systems in the presence of these different forms of uncertainty increases the computational cost considerably due to the iterative nature of these algorithms. Therefore, one of the objectives of this research was to develop methodologies that can efficiently handle multiple forms of uncertainty. Most of the work available in the literature about uncertainty analysis deals with the estimation of the safety of a structural component based on a particular performance criterion. Often an engineering system has multiple failure criteria, all of which are to be taken into consideration for estimating its safety. These failure criteria are often correlated, because they depend on the same uncertain variables and the accuracy of the estimations highly depend on the ability to model the joint failure surface. The evaluation of the failure criteria often requires computationally expensive finite element analysis or computational fluid dynamics simulations. Therefore, this work also focuses on using high fidelity models to efficiently estimate the safety levels based on multiple failure criteria. The use of high fidelity models to represent the limit-state functions (failure criteria) and the joint failure surface facilitates reduction in the computational cost involved, without significant loss of accuracy. The methodologies developed in this work can be used to propagate various types of uncertainties through systems with multiple nonlinear failure modes and can be used to reduce prototype testing during the early design process. In this research, fast Fourier transforms-based reliability estimation technique has been developed to estimate system reliability. The algorithm developed solves the convolution integral in parts over several disjoint regions spanning the entire design space to estimate the system reliability accurately. Moreover, transformation techniques for non-probabilistic variables are introduced and used to efficiently deal with mixed variable problems. The methodologies, developed in this research, to estimate the bounds of reliability are the first of their kind for a system subject to multiple forms of uncertainty