Hybrid engineering models — models that combine both analytic and heuristic knowledge — can address a wide range of design issues. However, models that combine these two types of information are often non-smooth and thus cannot be optimized by traditional derivative-based optimization algorithms. This paper is a study of four non-derivative optimization methods: Simulated Annealing, Genetic Algorithms, Flexible Polyhedron Search, and Function Approximation. Each algorithm was tested on two engineering problems that combine heuristic and analytic information. The algorithms were then evaluated with respect to ease of implementation, robustness, and efficiency.
Although each algorithm has advantages/disadvantages that may make it preferable for specific problems, in this study the Flexible Polyhedron Search was the best algorithm both in terms of ease of implementation and efficiency. This method found good designs in a relatively few number of design analyses. It was not, however, as robust as Simulated Annealing.