National Technical University of Athens (NTUA)
Two-time response excitation theory for non linear stochastic dynamical systems
Abstract
dc:descriptionIn the context of probabilistic modeling uncertain parameters/functions are quantified as random. Ingeneral, the characterization of random functions is a difficult task involving the knowledge of thehierarchy of the probability distributions of all orders or, equivalently, the knowledge of thecharacteristic functional. If random parameters/functions enter either as random initial conditions or asinput/excitation (external and/or parametric) to dynamical systems, then the systems output/responsewill also be a random function. In case that the dynamics can be modeled by means of differentialequations these equations are called Random Differential Equations (RDEs). The difficulty ofcalculating the probabilistic characteristics of the response is drastically reduced when we assume thatthe stochastic excitation is delta correlated. However, this assumption is not plausible when thecorrelation time of the excitation is of the same order of magnitude as the system’s relaxation time, asis the case for macroscopic dynamical systems, e.g. for systems excited by sea waves, wind loads, orearthquakes. In this case the excitation can be realistically modelled by smoothly correlated (colored)random functions. RDEs with colored excitation (also known as generalized Langevin equations)involve an increased amount of complexity due to the fact that in order to obtain system’s responseprobabilistic structure one has to consider infinite dimensional differential equations. Although thegeneral case of smoothly-correlated excitation is the most interesting case for many applications inengineering and applied sciences, existing methodologies fail to treat it in a satisfactory way.In response to this situation, the response-excitation (RE) theory, a new method for the probabilisticcharacterization of any non-linear system with any type of smoothly-correlated random excitation, hasbeen recently introduced by Athanassoulis & Sapsis (2006) and Sapsis &Athanassoulis (2008). TheRE theory, proposes the joint treatment of the probabilistic structure of the response and the excitation,leaving the space for their stochastic dependence to be determined during the solution of the problem.Athanassoulis and Sapsis used the characteristic functional approach to derive an equation for the jointRE characteristic functional and showed that, by appropriately projecting this infinite dimensionalequation, it is possible to obtain equations for the evolution of the joint response-excitation probabilitydensity function (REPDF). The derived joint REPDF evolution equation is a peculiar one, involvingtwo times (one for the excitation, s, and one for the response, t ), and partial derivatives only withrespect to one of them (response time), whereas, after the differentiation, the limit of the excitationtime st should be taken. I.e., the REPDF evolution equation includes the “half-time” derivative( ) ( ) ( ) / x t y s s tf t¶ a,b ¶ . This peculiarity gives rise to fundamental questions regarding both the wellposednessand the methods of its numerical solution. While working on this thesis, it became evidentthat the joint REPDF evolution equation of Athanassoulis and Sapsis in not a closed equation, and thuscannot provide a unique REPDF. The same finding has also been stated recently by Venturi et al(2012). This is due to the fact that when the half time limit ( ) ( ) ( ) / x t y s s tf t¶ a,b ¶ is considered thenon-local (in time) characteristics of the problem are partially lost. The present work continues thestudy of the RE theory, aiming at the clarification of various obscure points, and its furtherdevelopment towards the implementation of efficient algorithms for numerical solutions.In the first part of this thesis, the RE theory, introduced by Athanassoulis and Sapsis, is reviewed andgeneralized to second-order nonlinear systems. The joint REPDF evolution equation for non-lineardynamical systems under smoothly-correlated stochastic excitation is re-derived, using thecharacteristic functional approach. To verify the validity of the obtained equations, the latter have beenused to re-derive the infinite system of the limiting two-time moment equations, which are alsoobtained directly from the dynamical system. Finally the joint REPDF evolution equation is specifiedto the case of the ship roll problem.Subsequently, a well-studied, simple problem is considered in the context of the RE theory. Moreprecisely, the two-time RE moment equations are developed for a linear scalar dynamical systemunder colored stochastic excitation. These equations are solved analytically and results are obtainedfor different stochastic input functions. For Gaussian excitation, a complete analytical solution of thestudied problem, both in the transient and in the long-time statistical equilibrium state, is produced.The analytical solution of this simple problem is used in order to verify/clarify the REPDF evolution equation for linear RDEs, and prove that it can have multiple solutions. Thus, the need for an a prioriclosure of the REPDF evolution equation, providing additional information about the RE correlationstructure, becomes evident. The formulation and implementation of an efficient closure of this type isone of the fundamental contributions of this Thesis.The findings from the study of the linear/Gaussian case are generalized for the non-linear/non-Gaussian case, drawing, also, on evidence gained looking into Monte Carlo (MC) simulations results,performed by Z.G. Kapelonis. In fact, in the long-time statistical equilibrium state the joint REPDFtends to concentrate around the equilibrium curve of deterministic problems realized on the RE-phasespace. Reclaiming these findings, new, auxiliary, local conditions are developed in the RE-phasespace, by the use of local linearizations/Gaussianizations around the equilibrium curve of the nonlinearscalar dynamical system in the long-time. These, analytically solvable, local conditions, cansuccessfully approximate the local RE correlation structure as is verified by comparisons with resultsobtained by MC simulations and, therefore, can be used to form a new a priori closure scheme for thenon-linear REPDF evolution equation. The analytically obtained local closure information for the REcorrelation structure is synthesized in the REPDF evolution equation by the use of an appropriaterepresentation of the two-time joint REPDF, consisting of a superposition of Gaussian Kernels. Thereformulated REPDF evolution equation, together with the new local closure conditions, isnumerically solved using a Galerkin scheme. This allows for the specific structure of the consideredRDE to enter in the Galerkin coefficients both explicitly thought their dependence from the equationsof the dynamical system to be solved and implicitly through the Kernel coefficients which containinformation from the family of the localized problems. The Galerkin coefficients, having the form ofproducts of polynomials with bi-dimensional Gaussian densities are analytically calculated, and theproblem is solved as a constraint minimization problem. This Galerkin scheme has been used for thedetermination of the joint RE probabilistic characteristics of a half-oscillator, subject to asymptoticallystationary, colored, Gaussian or non-Gaussian (cubic Gaussian) excitation. The obtained results aresatisfactorily compared with solutions obtained from MC simulations for the same problem.The selection of the appropriate computational domain for the numerical solution of the joint REPDFevolution equation in the long-time, initiated the development of a new methodology for theformulation and solution of a system of two-time RE moment equations. These equations can apply toany non-linear system with arbitrary polynomial non-linearities, excited by colored Gaussian orpolynomially non-Gaussian processes. More precisely, moment equations for the response mean valuemx (t) , the two-time RE cross-covariance Cxy (t, s) , two-time response auto-covariance Cxx (t, s) andtime-diagonal response auto-covariance Cxx (t, t) are derived directly from the dynamical system. AGaussian closure condition is, then, applied in order to eliminate the higher order moments from thetwo-time moment equations. Following the Gaussian closure, considering s as a parameter, thederived equations can be considered as linear ODEs with respect to t , having coefficients dependedon the time-diagonal moments. The equation for Cxy (t, s) is used to express Cxy (t,t) as a non-linear,non-local in time (causal) operator on the whole history of mx (u) and Cxx (u , u) , t0 £ u £ t . Usingthe obtained operator for Cxy (t,t) , a closed, non-linear, causal system of evolution equations formx (t) ,Cxx (t, t) is obtained. After solving this causal system, the two-time moments can be calculatedfor all ( t , s ) pairs as well. Results obtained by the direct solution of the two-time RE momentequations in the long-time, statistical equilibrium limit are presented. Moreover, a first idea on a bi-Gaussian moment closure scheme that could extend the presented methodology to bi-stable halfoscillators in the long-time limit is discussed. Obtained results are compared with MC simulationssatisfactorily in the mono-stable case. In the bi-stable case the discussed bi-Gaussian moment closurescheme gives acceptable, preliminary, results only for the time-diagonal moments and the two-timeRE cross-correlation, whereas, in its present form, fails to approximate the two-time response autocorrelation
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2013
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsantili, Sevasti-Ivi
- Τσαντίλη, Σεβαστή-Ήβη
Subjects
dc:subject × 27- Στοχαστική μοντελοποίηση μη-γραμμικών συστημάτων
- Μη-γραμμικές τυχαίες διαφορικές εξισώσεις
- Ομαλά συσχετισμένη στοχαστική διέγερση
- Μη γκαουσιανή διέγερση
- Μη μακροβιανή απόκριση
- Θεωρία απόκρισης διέγερσης
- Μέθοδος του χαρακτηριστικού συναρτησιακού
- Εξίσωση εξέλιξης της από κοινού σππ απόκρισης και διέγερσης
- Εξισώσεις ροπών απόκρισης διέγερσης δύο χρόνων
- Μη τοπικές εξισώσεις - αιτιατοί τελεστές
- Stochastic modeling of non-linear systems
- Non-linear random differential equations colored stochastic excitation
- Non-gaussian excitation
- Non-markovian response
- Response-excitation theory
- Characteristic functional approach
- Joint response-excitation pdf evolution equation
- Two-time response-excitation moment equations
- Non-local equations-causal operators
- Φυσικές Επιστήμες
- Μαθηματικά
- Επιστήμες Μηχανικού και Τεχνολογία
- Άλλες Επιστήμες Μηχανικού και Τεχνολογίες
- Natural Sciences
- Mathematics
- Engineering and Technology
- Other Engineering and Technologies
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/35137