| A major goal of structural design is to achieve harmony between maximum safety and minimum cost. Conventional deterministic structural optimization has been able to accomplish such a goal over the last three decades. Due to the increasingly competitive structural design market, deterministic optimum designs are pushed to the limits of the constraint boundaries leaving no room for material, loads, modeling and performance uncertainties. Optimization under uncertainty serves the task of structural optimization and defining full statistical models of both loads and resistance. Reliability-based design optimization (RBDO), life cycle cost optimization (LCCO) and life cycle utility optimization (LCUO) procedures fall under the umbrella of the concepts of optimization under uncertainty. Lately, LCCO and LCUO have been seen as the most rational procedures that ensure long term benefits to society, owners and operators on infrastructural systems where the design is highly dependent on the failure consequences. Considerable research has been conducted to find a unified analytical and numerical procedure for solving problems of optimization under uncertainty. However, the developed methodologies are still in the academic envelope. Nowadays, the engineering and industrial communities recognize the importance of the direct involvement of uncertainty in their analyses and optimization. While its importance has been recognized, uncertainty has not yet been widely exploited. More simplified and practical methods that are both mathematically and numerically acceptable need to be developed.; In this thesis, decision making theory was employed to formulate LCUO/LCCO of facilities under infrequent hazards. Such hazards were idealized as renewal processes with a special case of a Poisson process. A new surrogate model for optimization under uncertainties was developed. The proposed surrogate model replaces the complicated constraints such as the probability of failure with a continuous and differentiable function. A comparison between the established approaches and the new approach was conducted in order to shed the light on the versatility of the proposed methodology. The structural response can be given in an explicit form and most of the time is implicitly evaluated through structural analysis tools. The surrogate model was developed for both explicit and implicit performance functions by adopting the traditional response surface method (RSM). An artificial neural network (ANN) was introduced to approximate implicit limit state functions. However, the main contribution of the ANN to the optimization under uncertainty process was to reduce the numerical effort required to form the approximate limit state functions. Because of the contradictory nature of safety and cost, there is no apparent universal rule of optimality for structural design. Therefore, multi-objective optimization (MOOP) was introduced to support the decision maker with a pool of optimal solutions. A simplified approach based on a hypo-ellipse approximation for the Pareto optimal front was introduced. The developed methodology and the simplified procedures were applied to two applications: transmission line design and the corrosion of pipelines were investigated.; Although the surrogate models are approximate, they have the advantage of convenience for industrial applications. The main conclusion drawn at the end of the thesis is that problems of optimization under uncertainty can be evaluated efficiently through numerical approximation. Therefore, designers are no longer hand-tied when it comes to the combination of optimization and reliability analyses. |
