Automatic machine learning versus human knowledge-based models, property-based models and the fatigue problem

Abstract

This paper is devoted to emphasizing the importance of human-based knowledge and to present the original property-based models. The main idea is that models are built in terms of equations that reproduce and guarantee their satisfaction, which leads to non-arbitrary parametric models. The methods based on data alone are not sufficient to be applied to fatigue S–N and GRV-N models: first, the required high number of results in machine learning is not attained in fatigue; second, a black box cannot supply a comprehension of the fatigue phenomenon; third, the absence of a supporting model impedes extrapolation beyond the scope of experimentation; and fourth, many other data are required to include the stress ratio, R , while robust models are already available. These models are artificial intelligence mixed models, where its main part is human-based whereas the parameter estimation is solved based on data. The fatigue problem is used to illustrate the methodology and show that its generalization is applicable to other real problems. A detailed analysis of the fatigue model properties is done to show the readers how to extend those properties to other cases. Some future lines of research are suggested, followed by some conclusions. This article is part of the theme issue ‘Physics-informed machine learning and its structural integrity applications (Part 2)’.