Tendon layout optimization in statically indeterminate structures using neural networks and genetic algorithm

Prestressing is a viable and effective technique used in concrete structures that allows the improvement of their performance. Its efficiency depends on the layout of the tendons, understood as the number of strands, their profile, curvature, and eccentricities. Layout design is a laborious and usually iterative trial-and-error process that aims to ensure that stress limits are not exceeded while, on the other hand, to minimize the total weight of prestressing steel. To automate the process, this research presents a novel genetic algorithm and a machine learning-based approach with a parametrically defined geometry of a tendon consisting of straight lines and circular arcs. The paper describes assumptions, implementation, and validation of the methods using a two-span two-girder bridge structure as an example. The genetic algorithm was incorporated to minimize the total weight of prestressing steel and ensure a correct design with no limit state exceedances. Since prestress evaluation in statically indeterminate structures requires repetitive and time-consuming numerical simulations, usually finite element analysis (FEA), a neural network was used to predict numerical output based on the layout parameters. The introduction of the neural network significantly reduced the total computation time. The results show the usability of the proposed approach in further studies and in engineering practice.