Evolutionary and GPU computing for topology optimization of structures

Although structural topology optimization, as a discrete optimization problem, has been successfully solved several times in the literature using evolutionary algorithms (EAs), the two key difficulties lie in generating geometrically feasible structures and handling a high computation time. These two challenges are addressed in this paper by adopting triangular representation for two-dimensional continuum structures, related crossover and mutation operators, and by performing computations in parallel on the graphics processing unit (GPU). Two case studies are solved on the GPU that show 5× of speedup over CPU implementation. The parametric study on the population size of EA shows that the approximate Pareto-optimal solutions can be evolved using a small population with the proposed EA operators.