Digital Methods for the Fatigue Assessment of Engineering Steels

Abstract

Engineering steels are used for a wide range of applications in which their fatigue behavior is a crucial design factor. The fatigue properties depend on various influencing factors such as chemical composition, heat treatment, surface properties, load parameters, microstructure, and others. During product development, various material characterization and qualification experiments are mandatory. For a faster and more cost‐efficient development, data driven methods (machine learning) promise to replace or to complement material testing by prediction of the fatigue strength. With an ontology‐based, semantically‐linked knowledge graph, representing the manufacturing history of the material, the influence of the parameters of the process chain on the resulting properties can be accounted for. Herein, it is shown how a fatigue database containing a wide range of materials is assembled from literature. After postprocessing and curation of the data, machine learning predictions of mechanical properties are discussed under multiple aspects. A domain ontology is defined, containing the relevant class definitions for the use case. After applying a data integration and mapping workflow, it is shown how the data can be systematically queried using knowledge graphs describing the manufacturing history of the materials.