Artificial Neural Network Model for Analysis of Linear Dynamic Systems Subject to Non-stationary Excitations

Pawarid Posayanant, Patiphan Chantarawichit, Damang Dy, Yos Sompornjaroensuk

Abstract

Computation of stochastic responses is a key step in reliability and probabilistic analysis ofdynamic systems. Monte Carlo Simulation (MCS) is generally employed for accurate analysis. ArtificialNeural Network (ANN) has a capability of mapping input to output. Due to the requirement of largesample sizes in the reliability and probabilistic analysis, ANN has been successfully applied as a surrogatemodel in many applications except non-stationary excitations. This paper proposes for the first time toapply ANN as the surrogate model for the non-stationary excitation. Specifically, multi-layer feed-forwardANN is employed for the purpose. The applicability of the proposed methodology is illustrated througha probabilistic analysis of a 3-DOF linear system subjecting to non-stationary ground excitation. Thenumerical results shown the potential of the proposed methodology.