Foley Much of the past research in optimization of structures has dealt with trusses, but recently partially and fully restrained steel frames have been optimized using classical as well as genetic algorithm procedures. However, optimal structure design procedures have been limited to those involving ASD and/or LRFD design specifications. Advanced analysis is a means with which proportioning of a steel frame is undertaken with a design analysis that includes all aspects of frame and member behavior. Such an analysis allows the LRFD design equations to be omitted. Therefore, optimal design utilizing advanced analysis offers substantial improvement for least cost designs, since the design analysis allows for full redistribution of forces within the frame and avoids excessive overstrength that can occur with member by member design procedures.
This research seeks to develop and test methods of implementing genetic algorithms in the optimal design of steel frames. Fully and partially restrained and flexible connections will be considered. An unconstrained optimization problem that includes provisions for advanced analysis in the constraint equations will be formulated. The computationally intensive distributed inelasticity procedure for computing nonlinear inelastic load-deformation response will be implemented using parallel processing and vectorization. A novel approach to limiting the search space using a genetic tree representation for structural steel frameworks will be used. This representation gives a more intuitive feel to problems and also suggests macro crossover operations that would be difficult to discover using traditional binary string representations. A method of intelligent mating using fitness proportionate sampling and normalized genetic differences with respect to cost and penalty functions will be implemented and evaluated. A cross-over frequency parameter is to be defined whereupon the amount of genetic material swapped during cross-over operations can be regulated. Parallel processing and vectorization will be implemented and evaluated in the optimization procedure. ***
