In real-world structural optimization, it is reasonable considering discrete design variables. Genetic algorithms would be suitable for these discrete structural optimization problems on account of their inherent features of discreteness and globality. But they generally require a large amount of computational efforts. In this thesis, an algorithm is proposed for the optimal design of discrete structures which is composed of a genetic algorithm and a local minimization method. The genetic algorithm provides good initial design points to the local optimizer, so convergence might be accelerated. A discrete neighbor optimal point is found by evaluating approximately all discrete design points neighboring the continuous solution which is obtained by the local minimization procedure. These discrete points are included into the population of the next generation, so design group may be composed with better chromosomes than before.
The proposed algorithm is applied to the design of truss and beam structures. The results are compared with those from other methods such as a genetic algorithm, a branch and bound method and a dual method. The present results show higher robustness and reliability. The present algorithm is much more efficient than genetic algorithms, but it still requires further improvement to be practical.
