Prediction of structural failure in single-story frame buildings under dynamic horizontal loading using Feed-Forward Neural Networks

In the presence of the evolution of computational mechanics and machine learning sciences, the formulation of estimation tools is of significant importance. In this article, the construction of Feed-Forward Neural Networks is presented for the estimation of failure load, failure time, and corresponding peak structure displacements, velocities and accelerations. The dataset was obtained from the computational nonlinear dynamic failure of a single-story building. It has been demonstrated that the supervised learning procedure has converged rapidly, at 4 epochs, and with a small mean squared error, namely about 1%. Moreover, the correlation between the outputs and targets of all subsets of the initial dataset vector is very strong. Following the model’s reliability, it has been proven that the presence of resonance is making significant amplifications to the response of the structure, in the order of magnitude of the theoretical value of the dynamic amplification factor of 10 for sinusoidal waves. Ultimately, the formulated Networks comply with the physical constraints and are reliable for estimations that will assist in to design of earthquake-resistant infrastructures.