EXPERIMENTAL AND ANALYTICAL STOCHASTIC FATIGUE OF WELDED STEEL JOINTS
Doctor of Philosophy
Thirty-nine welded steel specimens have been tested under variable amplitude loadings to determine the extent to which fatigue life is affected by the amplitude variability of the loading and by the high frequency components in the psd of the loading. Constant amplitude tests have also been run to determine the S-N curve so that comparisons can be made between the variable amplitude results and the commonly used analytical techniques such as rainflow analysis, the Rayleigh approximation and the peak counting method. Analytical predictions are based on an S-N curve corrected for the effect of weld residual stress. The twelve different variable amplitude loadings are classified into two groups. The variability of the amplitude of the group I loadings depends on the correlation between the successive extrema. The loadings vary from broadband to extremely narrowband, having peak to adjoining valley correlation values from zero to $-0.995$. Most of these loadings have Rayleigh distributed peaks and valleys. The group II loadings deal with the effect of high frequency components of psd for Gaussian processes. This includes variation of the relative frequency location of the high frequency component in a bimodal psd, and variation of the frequency of truncation of a smooth psd curve. The experimental fatigue lives for group I loadings are all significantly smaller than the predictions by either the Rayleigh approximation or rainflow analysis. The lowest fatigue life found is only 35% of the Rayleigh approximation. The experimental results for group II loadings indicate that the presence of a high frequency component can reduce the fatigue life substantially (as much as 46% for one case studied). Furthermore, it is found that neither of the commonly used analytical techniques can consistently predict the effect of high frequency components. A new damage model is also developed in an attempt to obtain predictions in agreement with the experimental results. The new theoretical results are in very good agreement with the experimental results for group I loadings and group II loadings with unimodal psd's. However, for the two group II loadings with bimodal psd's, the new damage model gives quite conservative predictions.