Computational Approaches for Decision Support in Structural Performance Evaluation

Structural engineers face complex decision making in the design and performance evaluation of structural systems. The complexity of the problem is generally determined by factors like continuous and/or discrete nature of its decision variables and presence of multiple objectives. Often, many solutions are available for the problem-at-hand and significant time is spent in identifying the final solution due to the iterative nature of the design process. Formulating this problem as an optimization problem has enabled the application of numerous approaches from operations research. Previously, these approaches were not applicable in most cases either due to the enormity of the decision space or the computational complexities associated with the design problem. The availability of processing power and memory as a result of recent advances in computing technology, has now evoked interest in the application of sophisticated optimization techniques like genetic algorithms for structural optimization. But, current research in structural optimization suffers from two major limitations that are related to the way optimization is currently employed for supporting the decision-making process. The first limitation is related to the adaptation of mathematical optimization techniques for application in structural engineering. Researchers have tended to use optimization techniques like black-boxes. It is possible to develop special adaptations to the mathematical optimization techniques for the particular problem-in-hand that may significantly improve their overall performance. An example of such an adaptation may be related to adapting Genetic Algorithms (GA) to problem-dependent implementation based upon knowledge of structural behavior. Another limitation of current optimization studies lies in their suggested use to find a single “optimal” solution. The final solution from an optimization approach is rarely the best solution because optimization formulations often exclude certain objectives and constraints due to the difficulty in quantifying them. One way of overcoming this drawback is by using optimization to generate a set of “good” alternatives. Then, the engineer will have the opportunity to study these alternatives, conduct what-if analyses, make minor changes if needed and pick the best solution from the set. The engineer can then employ professional judgment and expertise to evaluate these alternatives with respect to unmodeled factors. Many structural design problems belong to the class of discrete optimization. For example, the types of available products for various members in a truss are often chosen from a discrete 17 ANALYSIS AND COMPUTATION SPECIALTY CONFERENCE th