Figure 11 illustrates the PGA process used in this application. An initial population of 200 variants is created by random selection of design variables within the design space specified in Table 1. A chromosome with 20 design parameters as in Figure 10 represents each variant. The ship variants defined by these chromosomes are balanced, and evaluated using the ship synthesis model resulting in an assessment of feasibility and objective values (OMOE and LCC) for each variant. Fitness indicates a variant’s relative dominance in the population based on OMOE and LCC. A Goldberg (1989) ranking scheme is used. Variants are sorted into layers of Pareto-dominance. Each layer contains all variants that are dominant to subsequent layers. Variants are sorted with the best variant in the highest layer getting a rank of one and the poorest in the lowest layer a rank of 200 (population size). A geometrically decreasing probability of selection is assigned to each variant based on its rank. Equivalent variants (same dominance layer) are ordered randomly within their layer, probabilities are averaged for variants in the same layer, and the same average value is ultimately assigned to each. Variants are penalized for infeasibility and for similarity to other variants. This minimizes niches and duplicate variants, and forces the selection to spread out over the objective frontier.