Developments in Probability-Based Strain-Life Analysis

An initial statistical approach to defining a probabilistic strain-life curve has been developed for material fatigue strength characterization in the low-cycle region. It is based on the statistics of the standard linear regression model, and assumes that the scatter of the base 10 log of low-cycle fatigue life is normally distributed, with a Constant variance. Preliminary low-cycle fatigue data from a large-scale, experimental strain-life material test program were used to fit the probabilistic strain-life curve, using the probabilistic linear regression model derived herein. Upper and lower confidence bounds on the fatigue life prediction intervals were generated, with all of the test data falling within the 95 confidence region. The probabilistic strain-life model has the capability to be integrated into a Monte Carlo simulation of structural component fatigue life reliability analyses. The influence of two different ASTM low-cycle fatigue test specimen geometries on fatigue life scatter was also investigated. Crack growth in low-cycle fatigue tests, and the grain orientation of hourglass test specimens have both been shown to significantly affect the resulting scatter of fatigue life data at a given stress amplitude. The probabilistic linear regression model proves sufficient for characterizing the strain-life curve in the low-cycle fatigure region. More complex statistical methods that can account for nonconstant fatigue life variance, and fatigue life runout data, must be utilized to probabilistically characterize the straiin-life curve in the high-cycle fatigure region.