Frequency-based fatigue analysis of non-stationary switching random loads

The service loadings in real systems are not only random, but also non-stationary. The spectral methods based on a frequency-domain characterization of random loads, which have been used in alternative to classical time-domain approaches, cannot be applied to non-stationary loads, because the conventional spectral density spectrum is not able to capture the evolutionary frequency characteristics of non-stationary loads. This clearly restricts the applicability of the existing frequency-based methods only to loads which are stationary. At the same time, it is also very difficult to propose general models valid for all types of load non-stationarity encountered in practice. Therefore, a practical approach is to restrict the analysis to a specific class of non-stationary loads; in this work, we consider particular non-stationary loads (i.e. switching loads), which are piecewise stationary in their variance. A frequency-domain analysis of such loads is proposed, which is based on a combination of the frequency-based analysis of each adjacent stationary segment, which can be either Gaussian or non-Gaussian. Numerically simulated load histories, as well as loads measured on mountain bikes in special tracks, are analysed to validate the proposed methodology. The presented results also show the correlation between load non-stationarity and non-Gaussianity.