A simple linear regression strategy for fretting fatigue life estimates

R.A. Cardoso,G.A.B. Oliveira,G.M.J. Almeida,J.A. Araújo

DOI: https://doi.org/10.1016/j.triboint.2024.109852

IF: 5.62

2024-06-07

Tribology International

Abstract:Recently, an increasing amount of works have been considering the use of Machine Learning (ML) approaches to tackle the fretting fatigue problem. Notably, Artificial Neural Networks (ANN) models have been the most widely used ones. In this work, we propose a simple linear regression strategy to estimate fretting fatigue lives with a level of accuracy similar to that of more elaborate ML models. Such a simplification is possible by the judicious choice of the input parameters believed to govern the problem investigated. A fair amount of fretting fatigue data collected from the literature was used to test and validate the proposed methodology (206 tests). Life estimates by considering ANN models and a critical plane multiaxial fatigue criterion were used for the sake of comparison. Results have shown that, in terms of life estimates, the linear regression model here presented, in general, performed better than the multiaxial fatigue parameter considered and had an accuracy similar to that of the ANN models. The proposed regression model also provided accurate life estimates for challenging test conditions quite different from those used in its calibration. Moreover, compared to the ANN models, the linear regression one is considerably easier to implement, to train, and it allows an easier physical interpretation of the results.

engineering, mechanical