Fatigue crack growth prediction of automotive steels using ensemble-based machine learning methods

Abstract

The propagation of fatigue cracks under various loading conditions is one of the key problems in modern mechanical engineering and the operation of engineering structures. During operation, structural elements of machines and mechanisms are subjected to cyclic loading, which contributes to the initiation and gradual development of cracks. A crack is a source of stress concentration, and a local increase in stress occurs in the area of its tip, leading to its progressive growth. As a result, the cross-section of the part is gradually weakened, and failure occurs suddenly, without noticeable plastic deformation, even in highly ductile metals.

In this work, the fatigue life of automotive steel was predicted using the ensemble machine learning methods, namely, random forest and boosted trees. Experimental data describing the dependence of the fatigue crack length a on the number of loading cycles N and the overloading factor R_ol for the stress ratio R = 0.1 were used to train and test the models. The random forest method demonstrated high accuracy and stability of prediction with a mean absolute percentage error (MAPE) of 1.12%, while for boosted trees method showed the error equals to 3%. The results confirm the effectiveness of ensemble methods in predicting fatigue crack growth of automotive steels under cyclic loading conditions.