Multi-source data driven fatigue failure analysis and life prediction of pre-corroded aluminum–lithium alloy 2050-T8

Abstract

In this paper, the fatigue failure behavior of pre-corroded Al-Li alloy 2050-T8 was investigated through measured multi-source experimental information involving 3D corrosion morphology, 3D-DIC strain fields, SEM fractography, and EBSD material microstructure. A data-driven fatigue life prediction model for pre-corroded aluminum alloy was developed based on machine learning, in which the characteristic parameters of corrosion macro-morphology, local corrosion micro-morphology, and loading condition were trained by XGBoost algorithm to be associated with the residual fatigue life. Experimental results show that corrosion pits induce fatigue macro-crack initiation with varied time, amount, and specific location, due to the diverse corrosion morphologies. Four typical forms of fatigue micro-crack initiation were summarized, i.e., corrosion micro-pit, corrosion jut-in, corrosion tunnel, and pit sub-surface, mainly attributed to the mechanisms of local stress/strain concentration or local material embrittlement. It was demonstrated that the proposed data-driven model can effectively predict the fatigue life of pre-corroded aluminum alloy with high accuracy.