Random fatigue of structures with uncertain parameters and non-Gaussian stress response

Fatigue has been an active research topic in aeronautical engineering, civil engineering and mechanical engineering. The objective of this Ph.D. dissertation is to contribute to the further understanding of random fatigue with system uncertainties and non-Gaussian stress response.; We study the effect of structural parameter uncertainties on fatigue life under Gaussian random load. Structural uncertainties studied are from the geometry source, the material property source and the environmental source. Fatigue sensitivity to those uncertainties modeled as interval variables is investigated using the Monte Carlo simulation method.; We develop a new frequency domain procedure to predict fatigue life of structures with stationary non-Gaussian stress response. A probability density function (PDF) model for the rainflow range is proposed. The model captures a wide range of symmetrically distributed Gaussian and non-Gaussian processes characterized by four parameters of the process, namely, standard deviation, kurtosis, irregularity factor and mean frequency. The coefficients in the PDF model of rainflow range are determined using the multi-stage iterative regression method. A higher order moment constraint is used to improve the accuracy of the model. The explicit fatigue life formula obtained from the PDF model of rainflow range gives accurate predictions of fatigue life comparing with results from extensive Monte Carlo simulations.; We then extend the work to asymmetrically distributed non-Gaussian processes. A joint PDF model for the rainflow amplitude and mean is proposed as a function of the four characteristics chosen above and the skewness accounting for the asymmetry of the distribution of the process. The fatigue life predictions from the multi-stage iterative regression models agree well with those from extensive Monte Carlo simulations.; Finally we study the confidence interval for the prediction from the regression equations using the classical asymptotic approximation method. The 90% confidence interval is consistent with that of Monte Carlo simulations for stress processes with a range of irregularity factors.