Two-time, response-excitation moment equations for a cubic half-oscillator under Gaussian and cubic-Gaussian colored excitation. Part 1: The monostable case
Gerassimos A. Athanassoulis,Ivi C. Tsantili,Zacharias G. Kapelonis
DOI: https://doi.org/10.48550/arXiv.1304.2195
2013-04-08
Abstract:In this paper a new method is presented for the formulation and solution of two-time, response-excitation moment equations for a nonlinear dynamical system excited by colored, Gaussian or non-Gaussian processes. Starting from equations for the two-time moments (e.g. for Cxy(t,s), Cxx(t,s)), the method uses an exact time-closure condition, in addition to a Gaussian moment closure, in order to obtain a closed, non-local in time (causal) subsystem for the one-time (t=s) moments. After solving this causal system, the two-time moments can be calculated for all (t,s) pairs as well. The present method differs essentially from the classical Itô/FPK approach since it does not involve any specific assumptions regarding the correlation structure of the excitation. In the case where the input random process can be obtained as the solution of an Itô equation (as, e.g., happens with an Ornstein-Uhlenbeck process), the proposed non-local system is localized, leading to moment equations identical with the usual ones. The closed, non-local in time, moment system is numerically solved by means of an appropriate, two-scale, iterative scheme, and numerical results are presented for two families of colored stochastic excitations. The results are confirmed by means of extensive Monte Carlo simulations. It is found that both the correlation time and the details of the shape of the input random function affect appreciable the response covariance. In the present paper we focus on a monostable cubic half-oscillator, excited by a smoothly-correlated, linear-plus-cubic-Gaussian (non-Gaussian) random input. The bistable case, as well as more general nonlinear systems can be treated by the same method, provided that a more elaborate moment closure will be used.
Probability