Structural Reliability and Identification with Stochastic Simulation - Application to Railway Mechanics
Doctoral thesis, 2016

System identification of structures based on measured response data can play a key role in improving reliability based structural designs. However, the experimental limitations of in situ tests and uncertainties of required model complexity together with the inverse nature of system identification give rise to a number of challenging issues. Common examples of these issues in spatially varying parameter identification problems are poor modelling or ill-conditioning problems that are created due to inappropriate discretization of the parameter field. Even with a proper discretization, the large number of uncertain parameters associated with these problems makes the standard optimization or sampling techniques computationally cumbersome and also more prone to the so called curse of dimensionality problem. An in-depth study of such modelling and computational issues is presented for finding appropriate methods to treat them in a railway ballast stiffness-field property identification, and for doing test planning of in situ experimental conditions. This is achieved by utilizing a recently proposed Bayesian approach, known as enhanced Bayesian Updating with Structural reliability methods through feasibility studies. By interpreting the Bayesian system identification problem as an equivalent reliability problem, this approach opens up the possibility to employ well-developed rare-event samplers, such as subset simulation, to efficiently draw samples from the posterior probability distribution in high-dimensional inference problems. Another topic of this thesis is to develop a time integration scheme for fast simulation of large finite element models with spatially localized nonlinear or stochastic properties. This is a prerequisite for the ballast-sleeper load characterization problem, in which the local nonlinear/uncertain properties of the spatially varying ballast bed (along the sleeper length) is of major concern. Briefly stated, the developed integration scheme computes the system response based on the solution of an underlying linear system augmented with a low-rank nonlinear pseudo-force vector that accommodates the local nonlinearity and uncertainty effects. This is achieved through an efficient correction-prediction method. The presented integration scheme is combined with a developed modal reduction method, which is enhanced to take into account the effect of pseudo-forces in its modal dominancy analysis. It has been successfully applied to the studied moving-load simulation problem where the sleeper response statistics is required for estimating the risk of failure. The problem of detecting non-minimality in modal reduction of systems with multiple or very close eigenvalues (as in the studied railway track structure with large clusters of neighboring eigenvalues) is described and two methods to circumvent this problem is proposed. The reduction method is enhanced to effectively treat systems under moving loads or distributed loading, by incorporating information from structural properties of the input force matrix into the modal dominancy analysis.

model reduction

time-integration

finite element model

sleeper-ballast load characterization

stochastic simulation

Bayesian system identification

Room EA, Horsalsvägen 11, Johanneberg, Chalmers University of Technology
Opponent: Prof. Erik A. Johnson, Department of Civil & Environmental Engineering, University of Southern California, USA

Author

Sadegh Rahrovani

Dynamics

Subject Categories

Mechanical Engineering

Geotechnical Engineering

Infrastructure Engineering

Reliability and Maintenance

Probability Theory and Statistics

Areas of Advance

Transport

Infrastructure

C3SE (Chalmers Centre for Computational Science and Engineering)

ISBN

978-91-7597-348-7

Room EA, Horsalsvägen 11, Johanneberg, Chalmers University of Technology

Opponent: Prof. Erik A. Johnson, Department of Civil & Environmental Engineering, University of Southern California, USA

More information

Created

10/7/2017