This thesis aims to develop a software package that implements Canonical Variate Analysis in the MATLAB environment. Canonical Variate Analysis is a stochastic realisation technique that estimates the modal characteristics of dynamic systems from output-only vibrational data. The software will enable analysts to identify the natural frequencies, damping ratios and modeshapes of mechanical systems and structures while they are in-operation.
A two-tiered research approach was undertaken to address modal analysis in general as well as the theoretical background of Canonical Variate Analysis. The research drew a clear picture of the Canonical Variate Analysis algorithm, its supporting assumptions, and its role in the field of modal analysis.
A software package was developed in the MATLAB environment implementing Canonical Variate Analysis. The program is capable of accepting stochastic vibrational data and returning a set of modal properties. A graphical user interface was also created to control the base code.
The software package was tested using damped and undamped virtual data, as well as real data from the Swiss Z24 highway bridge. The performance of Canonical Variate Analysis was excellent with the virtual data, outperforming the Complex Exponential method in both speed and accuracy. Undamped natural frequency estimates were generated in just seconds that deviated less than 0.02% from the analytical solutions.
Canonical Variate Analysis also performed well under real data testing. It was possible to identify a number of system modes that correlated with the findings in a paper submitted to the International Modal Analysis Conference 1999. The testing proves that the Modal Analyser package works as planned. It also establishes Canonical Variate Analysis as a fast and effective means of stochastic system identification.