GA-Based Decision Support for Optimizing the Response of Secondary Systems

Abstract

The objective of this research is to develop a Decision Support System (DSS) for seismic design and performance evaluation of piping supports. The current practice of designing piping support locations is primarily heuristic, relying heavily on professional experience. In this thesis, approaches to optimize support locations using Genetic Algorithms (GAs), a heuristic optimization technique, are discussed. These approaches have been implemented in a DSS using Vitri, a generic, distributed framework designed to support the development of DSSs, which reduces the computational requirements by combining the processing power of a network of workstations.

Previous attempts to solve the problem of pipe support optimization modeled supports as flexible springs, which have a stiffness depending on the support capacity, resulting in the use of an integer representation in the GA. In this thesis, a new approach where supports are modeled as rigid springs is presented. This permits the use of a binary representation in the GAs. Also, earlier attempts had solved the problem by minimizing the number of supports, which does not always indicate if cost is minimized. In this thesis, capital cost and lifetime cost are studied by examining the trade-off curve between the cost and the number of supports. A crossover scheme aimed at generating cost optimal solutions of a specified number of supports, which is required for generating trade-off curves, is proposed.

It has been observed that optimization results in solutions that may be practically infeasible because of unmodeled costs in the optimization model. In pipe support optimization, such costs might be from the preference of certain locations over others because of easier support installation costs or the desire to locate the supports under lumped masses to stabilize the pipe against local vibrations from equipment such as pumps and motors. The role of Modeling to Generate Alternatives (MGA), a methodology based on optimization to produce alternatives, is explored to address these issues.