Extreme response and fatigue damage analysis of wind-excited structures considering non-gaussian load effects
MetadataShow full item record
Probabilistic assessment of the performance of structures subject to natural and man-made loads is widely recognized as a rational way in structural design and risk mitigation, in which an important problem is to determine the probability of structural response exceeds some specified limit states, say the limit states related to first-passage or fatigue failure. This is a challenging problem in the theory of random vibration and reliability analysis and has attracted enormous amount of attention. For wind-excited structures which is focused in this study however, there is still no general solution when the structural system is complicated and the failure probability is small, especially when the non-Gaussian wind load effects is considered. In this dissertation, an effective controlled Monte Carlo simulation (MCS) framework is firstly proposed for estimating small failure probabilities of dynamic structures. It combines the importance splitting (ISp) method with multivariate autoregressive (MAR) modeling of stochastic excitations. The ISp method, also referred to as subset simulation with splitting, splits important sample paths into multiple branches at various stages in the simulation. It permits the estimation of a small failure probability of a rare event through estimations of conditional probabilities of intermediate events. The MAR model of excitations is established based on their spectral matrix, which transfers the stochastic excitations as the output of a loading system with uncorrelated white noise process as input. This scheme is very efficient in generating offsprings of loading and response time histories conditional on the intermediate events with very low rejection rate, which facilities the application of ISp method to different kinds of stochastic excitations. Prediction of extreme responses with given mean recurrence intervals (MRIs) is an essential task for reliability-based design of wind turbines. Current turbine standards require a statistical extrapolation of short-term extreme response distribution at very upper tail region for the estimation of long-term extreme responses. However, the predictions from different approaches are significantly different. The proposed controlled MCS enables a direct simulation of long-term extreme responses with much reduced computational efforts, the extreme value distribution determined therefore provides a valuable opportunity to verify the adequacy of different statistical extrapolation approaches. New insights and improvements are provided in probabilistic modeling of wind turbine extreme responses. A comprehensive assessment of methods for extreme value analysis of non-Gaussian wind load effects using short-term time history samples is also presented. The methods examined are peaks-over-threshold method, the average conditional exceedance rate method, and the translation process method with various translation models. The long-term wind pressure data on a saddle-shaped large-span roof collected from wind tunnel test are used as the basis for comparison. These pressure data are featured by a variety of non-Gaussian characteristics, including mildly and strongly softening and hardening non-Gaussian processes with unique distributions. Some new developments of the methods are also presented to better predict the extreme value distribution taking into account the non-Gaussian characteristics. For translation process method, a new and theoretical sound moment-based translation model is derived for hardening non-Gaussian processes where the widely used Hermite model cannot be applied. The closed-form formulations for model coefficients in terms of process skewness and kurtosis are also presented. For fatigue damage evaluation, a recently proposed spectral method is re-evaluated for broad-band Gaussian and non-Gaussian wind load effects. The wind load effects considered are alongwind, crosswind and their coupled responses of tall buildings, and wind pressures on claddings. Following this spectral method, the rainflow counting damage is approximated by a linear combination of its upper and lower bounds. A new formulation for determining the combination factor is proposed which is in terms of process bandwidth parameters. The influence of sampling frequency on broad-band fatigue damage is also studied. For non-Gaussian wind load effects, research emphasis is placed on the modeling of translation function which is essential for non-Gaussian fatigue damage evaluation. The results demonstrate the effectiveness and accuracy of the spectral method for broad-band Gaussian and non-Gaussian processes through comparison with those from time domain rainflow counting method.