Statistical modelling of strain-life fatigue data: prior distributions based on estimation of cyclic properties

D. Leonetti ¹, J. Maljaars ^{1, 2}

¹ Eindhoven University of Technology, the Netherlands
² TNO, the Netherlands

The correlation between the cyclic strain-life parameters of the Basquin-Manson-Coffin relation and the monotonic strength properties of steels attracted the interest of researchers in the past decades. Their use is very convenient prior to the design stage, e.g. for material selection, since it allows a simple estimation of strain-life curves. If tests are performed and their results are analyzed according to standards, these correlations are of no use. Instead, the information contained in these correlations can be used as prior knowledge in a Bayesian framework if it is accurately formalized as a probability function, potentially leading to a reduction of the number of tests to be executed.

The present article formulates a probabilistic strain-life model based on the Basquin-Manson-Coffin relation and presents a statistical re-evaluation of two of the correlations between monotonic and cyclic strength properties proposed in the literature. Clear distinction is made between the intrinsic variability of the fatigue life and the epistemic uncertainty due to the number of observations. A joint multivariate probability density function of the cyclic strength parameters is formulated including both sources of uncertainty, which can be used in Bayesian analysis.

HERON
Home Back issues Search Information for authors About HERON	To the contents of Vol. 70, issue 2 To full-text article (pdf 288 KB)
	Statistical modelling of strain-life fatigue data: prior distributions based on estimation of cyclic properties D. Leonetti ¹, J. Maljaars ^{1, 2} ¹ Eindhoven University of Technology, the Netherlands ² TNO, the Netherlands The correlation between the cyclic strain-life parameters of the Basquin-Manson-Coffin relation and the monotonic strength properties of steels attracted the interest of researchers in the past decades. Their use is very convenient prior to the design stage, e.g. for material selection, since it allows a simple estimation of strain-life curves. If tests are performed and their results are analyzed according to standards, these correlations are of no use. Instead, the information contained in these correlations can be used as prior knowledge in a Bayesian framework if it is accurately formalized as a probability function, potentially leading to a reduction of the number of tests to be executed. The present article formulates a probabilistic strain-life model based on the Basquin-Manson-Coffin relation and presents a statistical re-evaluation of two of the correlations between monotonic and cyclic strength properties proposed in the literature. Clear distinction is made between the intrinsic variability of the fatigue life and the epistemic uncertainty due to the number of observations. A joint multivariate probability density function of the cyclic strength parameters is formulated including both sources of uncertainty, which can be used in Bayesian analysis.