| Popis: |
This PhD thesis studies the problem of online output-only stochastic structural system identification by recursive Bayesian estimators. To this end, Recursive Principal Component Analysis (RPCA) is used for model order reduction to speed up the computations. Several novel algorithmic deviations are proposed in the realms of RPCA, linear Bayesian smoothing, and nonlinear Bayesian filtering. The novel RPCA proposed in this PhD thesis outperforms the existing algorithms in computational speed, observability, and robustness. The novel smoothing method proposed in this PhD thesis notably reduces the noise of estimation and substantially improves the speed of cross-covariance calculations. The new nonlinear Bayesian filter proposed in this PhD thesis enables input-state-parameter estimation for systems without direct feedthrough without input regularization. |
