PSI - Issue 79

Lorenzo Leonetti et al. / Procedia Structural Integrity 79 (2026) 485–492

486

1. Introduction Advanced composites have progressively replaced traditional ones across a wide range of engineering applications, primarily due to stringent mechanical requirements (Barretta et al., 2015b, 2015a; Bruno et al., 2013, 2009, 2007; Luciano and Willis, 2005; Surjadi et al., 2019). In particular, bioinspired composites have recently emerged as the most promising for achieving extremely high performances due to their combination of stiffness, strength and toughness (Liu et al., 2020). These composites are regarded as highly heterogeneous and anisotropic, making their design and assessment particularly challenging, especially in relation to complex failure mechanisms. Consequently, accurate modelling approaches and efficient numerical solution strategies have become indispensable tools for their optimal design, especially when dealing with nonlinear dynamic analyses (Ammendolea et al., 2023, 2021; Askarinejad et al., 2018; Greco et al., 2018; Hu et al., 2014; Lonetti and Pascuzzo, 2016; Pascuzzo et al., 2022; Slesarenko et al., 2017). The most versatile modeling approaches for damaging composites belong to the class of multiscale models, by virtue of the optimal balance between numerical accuracy and computational efficiency (Canal et al., 2012; De Maio et al., 2025, 2024b, 2024a; Ghosh et al., 2001; Greco et al., 2021, 2017, 2015; Huang et al., 2019; Nguyen et al., 2012; Trovalusci et al., 2014). In recent years, reduced order models (ROMs) and data driven homogenization approaches have established themselves as popular surrogates of highly costly micromechanical models (Chafia et al., 2025; Logarzo et al., 2021; Oskay and Fish, 2007). This work aims to propose a novel, efficient data-driven multiscale strategy for the failure analysis of anisotropic composite materials, being able to overcome the limited efficiency of the existing multiscale models. This strategy integrates a homogenized continuum damage mechanics approach with a deep neural network trained by using micromechanical data obtained for the same microstructure subjected to several proportional macrostrain paths. As the main numerical application of this work, the proposed data-driven multiscale approach is used for analyzing the complex damage phenomena in nacre-like structures, modeled as staggered composites with periodically arranged platelets and cohesive joints placed in between them. In particular, several proportional and nonproportional macrostrain histories have been considered in the validation step, including loading-unloading branches to assess the capability of the proposed approach to comply with the damage irreversibility condition. 2. Multiscale modeling of staggered periodic composites 2.1. Simplified micro-modeling At microscale, staggered composites are modeled as periodic heterogeneous media as shown in Fig. 1. Units are made of a linearly elastic material, whereas matrix joints are lumped into zero-thickness cohesive interfaces with an isotropic damage evolution law (De Maio et al., 2019). In particular, the following constitutive relation is assumed for these interfaces: ( ) coh coh 0 1 0        = −         s s s n n n K t d K t   , (1) n t and coh s t are the normal and tangential traction components, respectively, n  and s  are the normal and where coh

n K and

s K are the normal and tangential stiffness coefficients,

tangential separation components, respectively,

respectively, and d denotes the damage variable, having the following expression:

0

if

max   0 

 

.

(2)

  

   

d

=

max       0 1 max 1 0    −

− 

min    

,1 else 

 



In Eq. (2) 0  and 1  represent the interface separations at damage onset and complete failure, respectively, and max  is the maximum value of the effective separation over the entire loading history.