Seismic pre-dimensioning of irregular concrete frame structures: Mathematical formulation and implementation of a learn-heuristic algorithm

Irregular frame structures can exhibit an inefficient dynamic behavior, such as undesirable torsional effects and asymmetric displacements, that can severely affect structural performances in case of a seismic event. To mitigate the influence of geometric irregularities, the pre-dimensioning of structural elements is crucial to achieve the best possible seismic behavior. Identifying the optimal solutions in terms of elements’ cross-sections and orientations is a non-trivial problem that in general designers tackle following trial-and-error attempts which can become numerous in the most complicated cases. This paper formally describes the pre-dimensioning problem and presents a Learn-heuristic that hybridizes a Genetic Algorithm and a k-Means algorithm procedure for its solution. The algorithm is validated on a diverse set of 3D reinforced concrete structures and through a Structural Pre-dimensioning challenge, where the performances of the algorithm are compared against those accomplished by a group of Ph.D. candidates. The computational results evidenced that the proposed optimization framework achieves pre-dimensioning solutions that significantly improve the designs provided by the Ph.D. students, thus suggesting a favorable inclusion of the proposed solution method in classical pre-dimensioning processes.