Contributions to the analysis of prestressed steel and concrete composite beams using machine learning algorithms

Resumo

In the practice of civil construction, one of the structural alternatives used so that beams can withstand the required loads in a more efficient and economical way are the prestressed composite beams, especially steel and reinforced concrete. Posttensioned composite steel-concrete beams (PSCCB) have a greater range of elastic behavior, as well as yield and ultimate load values. In continuous beams, there is a reduction of cracking in the hogging moment region. They may also have better fatigue performance and employ lighter steel sections. The present paper aims to create a large database for the analysis of PSCCB, based on a previously developed nonlinear finite element model, which performs static non-linear analysis of PSCCBs, considering the partial interaction between steel and concrete. With a database developed from a variation of the parameters in the numerical models, three Machine Learning models are developed and trained with the objective of predicting the beam ultimate load, deflection, and final tendon force. The implemented procedures is compared, whenever possible, with experimental and numerical results available in classic literature, mainly as a reference of test data to evaluate the success of Machine Learning algorithms.