Probabilistic Prediction Synthesizing Data and Science-Based Modeling for Very High Cycle Fatigue

Abstract

Abstract Epistemic and aleatoric uncertainty are inherent in engineering modeling and experimentation. Uncertainty, which must be considered for accurate life estimation and prediction, is an undeniable aspect of very high cycle fatigue (VHCF). Frequently, limited data are available for VHCF. The magnitude and extent of uncertainty are increased for high reliability applications or gigacycle predictions. The purpose herein is to present a methodology for managing uncertainty satisfactorily. The procedure integrates data with science-based modeling (SBM) for VHCF. An extensive set of data for SUJ2 steel, which exhibits bimodal damage growth for some loading conditions, will be used. One mode is associated with damage nucleating from internal particles, and the other is surface-induced damage, both of which can have fatigue lives in excess of 108cycles. One important conclusion is that synthesis of SBM with data greatly improves reliability estimation and prediction. As the accuracy of the SBM increases, the effect of uncertainty is reduced; however, even basic approximations are more beneficial than empirical analysis alone. Consequently, an SBM that reasonably predicts VHCF behavior resulting from multiple modes of damage growth, which is tuned subsequently by infusing VHCF data, is warranted. The approach focuses on the estimation and prediction of the cumulative distribution functions for life given the loading condition and their statistical goodness of fit.