Optimal design of steel building structures

Abstract

An optimization method has been developed to find the minimum weight design of the steel building structures which consist of the commercially available discrete section sizes. In this study, the emphasis was particularly placed on the practical applicability of the developed algorithm in the design practice of steel building structures which characterized by a simple objective function(weight of total structure) and a large number of design variables(sections) in the optimal design problem. In this study, the constraints imposed are decomposed into two levels(categories), namely, element level constraints and structural level constraints, and applied at the different stages of optimization process. The structure is optimized through element optimization under the element level constraints first and then, if there is any violation of the structural level constraints, it is adequately compensated by the constraint error correction vector obtained through the sensitivity analysis of the gradient projection method. On the other hand, in the discrete optimum design, occasionally, the solutions go back and forth repeatedly between the feasible and the infeasible regions during the series of redesigns of members with discrete sections. This oscillation phenomenon may be caused inherently by the discontinuity of variables of commercially available sections in the database. In this study, in depth investigation into the oscillation phenomena in the discrete optimization process and its control has been conducted. Additionally, there is a unique characteristic in the construction of the database. For one example, by dividing the available H-sections into several groups based on their section characteristics, much improved relationships between section variables were obtained and used efficiently in searching the optimum section in the section table. Besides this section, some sections and materials vigorously used in design practice, e.g., L-sections (angle), U-section...