The present work shows the analysis of a numerical experimental test that was performed using randomized and combined data to study the bending behavior of beams, through the deflection equation considering plane stresses in two different examples. The first is a cantilever beam with force concentrated at the free end and the second is a pinned-pinned beam with loading uniformly distributed along the span. The variables width, height, length, longitudinal modulus of elasticity, deflection, and loading were used as estimated parameters to calculate the ideal width and height dimensions for each beam and obtain a structural optimization considering the limits of deformation according to ABNT NBR 6118/2014. The data generation was generated in Excel spreadsheet format and worked in an Artificial Neural Networks in TensorFlow Python language, with six hidden layers. In addition, the functions 'mae', 'sgd' and 'loss' were used as optimizers or activation function in TensorFlow.