Issue 78

C. Popa et alii, Fracture and Structural Integrity, 78 (2026) 118-128; DOI: 10.3221/IGF-ESIS.78.07

deposition process generates an anisotropic structure in which the bonding quality between adjacent filaments and successive layers significantly affects the overall mechanical performance of the printed parts [1,4,6,7]. Several studies have demonstrated that raster orientation has a major influence on the mechanical response of FDM specimens. Ahn et al. [1] showed that ABS components manufactured by FDM exhibit pronounced anisotropic behavior depending on raster angle and air gap configuration. Chacón et al. [5] reported that the variation of raster orientation and layer thickness significantly modifies the tensile and flexural properties of PLA specimens. Similar conclusions were presented by Rankouhi et al. [8], who observed that failure mechanisms and crack propagation strongly depend on filament orientation and layer arrangement. Furthermore, Domingo-Espin et al. [9] experimentally and numerically characterized polycarbonate specimens produced by FDM, demonstrating the importance of process parameters in the prediction of the elastic response. In recent years, increasing attention has been focused on the shear behavior of FDM materials, since shear loading is strongly dependent on interlayer adhesion and filament arrangement. Torres et al. [4] investigated the optimization of PLA specimens subjected to shear loading and reported significant variations in shear strength depending on printing parameters. Additional studies showed that raster orientations such as 0°, 45°, and 90° lead to distinct stiffness values and failure mechanisms under shear-dominated loading conditions [3,8]. These observations confirm that the mechanical response of FDM structures cannot be considered isotropic and requires detailed experimental and numerical characterization. Besides raster orientation, the influence of perimeter shells or contour regions, investigated by Popa et al. [10], has also become an important research topic. The external shell may improve structural rigidity, reduce stress concentrations, and modify crack initiation and propagation mechanisms [2,3]. Cantrell et al. [7] demonstrated that contour configuration and filament deposition strategy influence the mechanical properties of ABS and polycarbonate printed structures. Similar trends were identified in studies focused on fracture and failure behavior, where contour regions contributed to changes in ductility and fracture energy [8]. These effects become particularly relevant for shear-loaded specimens, where the bonding between adjacent rasters and perimeter regions directly affects the stress distribution and crack evolution. Finite Element Analysis (FEA) is increasingly used to investigate the mechanical response of FDM structures and reduce the number of experimental tests required for material characterization. Numerical approaches range from homogenized orthotropic material models to detailed mesostructural simulations where individual filaments are explicitly represented [2,3,9]. Garzon-Hernandez et al. [2] proposed an experimental–numerical methodology for predicting the mechanical properties of FDM structures using mesostructural modeling, while Gonabadi et al. [3] demonstrated that microstructural finite element modeling combined with homogenization techniques can accurately predict the elastic response of FDM parts under tensile, bending, and shear loading conditions. These studies showed that numerical simulations are capable of reproducing the anisotropic behavior induced by filament orientation and deposition strategy. The primary standard for shear testing was proposed for thin sheet metals ASTM B831 [11]. For plastic sheets, ASTM D732 [12] is used by punch tool. These standards allow the determination of the material's ultimate shear strength. The previous attempts [10, 13] to determine the shear strength of additive manufactured materials used the methodology proposed in [11]. Artificial Intelligence (AI) and Machine Learning (ML) techniques have become increasingly important in engineering design and optimization due to their ability to identify complex relationships within large datasets and generate accurate predictions. In shape optimization problems, predictive models can significantly reduce the time and computational resources required for traditional experimental or numerical investigations [14,15]. We developed an AI-based methodology for optimal shape prediction using a supervised ML approach. This methodology consists of three main stages: data preparation and feature selection, preprocessing and dataset partitioning, and model development using Multiple Linear Regression (MLR). By learning the relationship between selected input parameters and the target shape response, the model provides a fast, interpretable, and computationally efficient framework for predicting optimal shape configurations and exploring new design alternatives without the need for additional simulations or experiments. Nowadays there are many attempts to use ML in data prediction of mechanical properties in order to save costly and timely experiment [16]. The objective of the study is to optimize the geometry of shear test specimens. The present study proposed a new shear specimen design based on numerical simulation and machine learning to find the optimal design of the shear specimens. The novelty of the present study lies in the integration of finite element simulations with a machine learning approach to optimize the geometry of shear test specimens. While previous works primarily focus on material behavior and process parameters, limited attention has been given to specimen geometry optimization for shear-dominated loading. The objective of this work is to determine the optimal combination of geometric parameters that maximizes the ratio between maximum shear stress and maximum normal stress (S =  max /  max ), ensuring a stress state closer to pure shear.

119