A meta-heuristic for topology optimization using probabilistic learning

Topological shape optimization consists in finding the optimal shape of a mechanical structure by means of a process for removing or inserting new holes and structural elements, that is to say, using a process which produces topological changes. This article introduces a method for automated topological optimization via an Estimation of Distribution Algorithm (EDA), a global optimization meta-heuristic based on probabilistic learning, which requires of neither user initialization, nor a priori information or design bias in the algorithm. We propose a representation and solution mapping which favors feasible structures and requires a, relatively, low dimensionality (some hundreds), the probabilistic model learns from finite element evaluations to generate well-performed structures. The EDA for topology optimization (EDATOP) is compared with an algorithm, in the state of the art, specially designed to address this problem, demonstrating that our approach is useful for solving real-world problems, escapes from local optima, and delivers better solutions than the comparing algorithm which uses problem knowledge, with a payoff on the computational cost.