Output-only measurement-based parameter identification of dynamic systems subjected to random load processes

In the present work a new output-only measurement based method is proposed which allows identifying the modal parameters of structures subjected to natural loads such as wind, ocean waves, traffic or human walk. The focus lies on the dynamic excitation of structures by wind turbulences and wind-induced ocean waves modeled as stationary Gaussian random process. In contrast to the existing output-only identification techniques which model the unmeasured load as white noise process, statistical information about the dynamic excitation, e. g. obtained by measurements of the wind fluctuations in the vicinity of the structure, are taken into account which improve the identification results as well as allow identifying the unmeasured load process exciting the structure. The identification problem is solved on basis of a recently developed method called H-fractional spectral moment (H-FSM) decomposition of the transfer function H(?) which allows representing Gaussian random processes with known power spectral density (PSD) function as output of a linear fractional differential equation with white noise input. In the present work the efficiency and accuracy of this method is improved by the use of an alternative fractional operator and a modification is proposed which makes it applicable to short as well as long memory processes. The most widely used wind and ocean wave model spectra are compared and discussed, and the corresponding H-FSMs are provided in closed form allowing to simulate realization of the processes in a straight forward manner. Based on the FSM decomposition a state space representation of arbitrarily correlated Gaussian processes is developed in closed form which neither requires the factorization of the PSD function nor any optimization procedure. Combined with the state space model of the structure, it leads to an overall model with white noise input, which can be efficiently combined with any state-space model-based parameter identification algorithms such as the well known (weighted) extended Kalman filter algorithm used here. The method is successfully applied for the stiffness and damping estimation of single and multi-degree of freedom systems subjected to wind and wind-wave turbulences as well as for the estimation of the unmeasured load process. Finally, a sensitivity analysis of the filter accuracy is conducted in order to improve the accuracy and efficiency of the method.