Capturing uncertainty in fatigue life data

摘要

Life data are investigated across a wide range of scientific disciplines and collected mainly through planned experiments with the main objective of predicting performance in service conditions. Fatigue life data are based on time, as measured in cycles, until complete fracture of a material in response to a cyclical loading. Fatigue life data have large variation, or dispersion, which is often overlooked or not rigorously investigated when developing predictive life models for the material. This research develops a statistical model of dispersion in fatigue life data which can then be used to augment deterministic life models yielding probabilistic life models. A predictive life model is developed using failure-time regression methods, to investigate the dispersion model. The predictive life and dispersion models are investigated as a dual-response function.