Infrastructure Engineering - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemModal testing of bridge superstructures subjected to traffic excitation using the methods of data dependent systems( 2003)The aim of this thesis covers the development and application of the time domain multivariate auto regressive moving average (ARMA) modelling approach to the system identification of bridge superstructures subjected to ambient traffic excitation. Procedures for system identification of structures can be divided into two categories: i) Frequency domain methods; ii) Time domain methods. Traditional frequency domain methods of system identification are primarily based on Fourier Transform techniques. The Fourier Transform of uniformly sampled data is usually performed using the so-called "Fast Fourier Transform" (FFT) algorithm. FFT algorithms are at the heart of most popular commercial Fourier analysers. The reason for their popularity is speed of execution and reliability. However, deficiencies exist in these methods, particularly when dealing with systems subjected to ambient excitation, high noise levels, high damping levels or closely spaced modes. The recognition of these specific weaknesses in the frequency domain approach coupled with the additional expense associated with the necessity of introducing controlled excitation of the structures under test has led to the development of several time domain based methods of vibration analysis. However, the implementation of multivariate ARMA modelling in structural system identification applications is demanding of computational resources. In addition, this form of modelling is complex requiring a user-friendly interface for it to be attractive to civil engineers. Recent developments in both computer and the advent of suitable software platforms such as MATLAB have made this task feasible for large civil engineering structures. The majority of civil engineers still prefer to use the FFT based computer packages which are readily available. The use of the multivariate ARMA model, however, is reasonably wide spread in other areas such as electrical engineering and economics, improving considerably on the trial and error approaches associated with most other techniques in this field. The general purpose of system identification of civil engineering structures is to serve as a tool for performing modal analysis. A major part of the study associated with this thesis has focused on the development of the mathematical framework of ARMAV models or equivalent stochastic state space realisation methodology for system identification of civil engineering structures subjected to ambient excitation. The other major part focused on the application of the developed mathematical framework to a simulated three degree-of-freedom system and to two different in-situ bridge superstructures subjected to ambient traffic excitation. For improving accuracy, the off-line non-linear Prediction Error Method (PEM) has been used for extracting model parameters and estimating covariance matrices. Also by using the PEM, it has been possible to estimate the standard deviations associated with modal parameter estimates. The results obtained from application of the ARMAV approach to in-situ bridge superstructures are compared with those obtained using the FFT-based Simplified Experimental Modal Analysis (SEMA) technique and the time-domain Random Decrement method (RANDEC). A finite element method (FEM) model was constructed for each application which was then updated on the basis of the bridge geometry and experimental results obtained after testing. It is concluded that the autoregressive moving average vector (ARMAV) model is a viable alternative to traditional FFT-based techniques especially in situations where modes are closely spaced and when data is highly contaminated with noise as it is generally the case with traffic excitation. In addition the ARMAV modelling approach when used together with the off-line non-linear PEM offers a means of identifying and estimating the errors in parameters estimated from the modelling approach (viz: natural frequencies, damping and mode shapes in the context of modal analysis) which is difficult if not nigh impossible to achieve using traditional Fourier based techniques.