Abstract

The presence of process-dependent, largely variable complex patterns of intrinsic defects such as lack of fusion or multiple interacting dispersed pores and voids has a major detrimental effect in the fatigue strength of additive manufacturing (AM) parts. These defects can trigger the nucleation and ultimate coalescence of small micro-cracks in airframe components subjected to vibratory environments, thus limiting the general implementation and certification of AM technology in the aerospace sector. Nowadays, industry efforts to consolidate a reliable behavioural modeling of AM components focus on three main areas: (i) collection of physical tests results to build comprehensive databases, (ii) statistical data treatment to enable stochastic approaches for AM materials and (iii) development of deterministic, high-fidelity numerical models for Virtual Fatigue Testing (ViFT). In this context, the characterization of AM micro-defects through surrogate models generated with Machine Learning (ML) algorithms is also arising as a keystone to reduce the volume of physical and numerical experiments needed to mimic the behaviour of AM parts at the different length scales of the test pyramid. This work presents ViFT simulations based on the Continuum Damage Mechanics (CDM) formulation at the defect and coupon scales. Two computational strategies, Finite Element (FE) and Element-Free Galerkin (EFG) methods, are benchmarked in problems with characteristic AM defectology. Different fatigue damage models, including the ones developed by Chaboche and Peerlings, are calibrated with experimental stress-life data in the High Cycle Fatigue (HCF) regime and coded into material subroutines. A CDM-meshfree EFG approximation with structured nodal arrangement and decoupled cell integration scheme is studied in order to assess alternative constitutive-discretization strategies for ViFT simulation at upper length scale levels of the test pyramid. Finally, a preliminary insight into the role of ML-based surrogate models in numerical simulations and their potential to tackle the sources of uncertainty of probabilistic ViFT is discussed.