Machine learning enhanced multi-scale dynamic viscoelastic analysis of 3-D printable PETG nanocomposite filaments: Leveraging FFT-based mesh-free computational homogenization for complex microstructures

Abstract

The article investigates the influence of organically modified montmorillonite nanoclay (OMMT-NC) and short carbon fibers (SCF) on temperature-dependent mechanical properties of additively manufactured glycol-modified polyethylene terephthalate (PETG) nanocomposites. This work utilizes Dynamic Mechanical Analysis (DMA) to explore the influence of microstructure on the multiscale viscoelastic properties and the resulting stiffness-damping trade-off in porous nanocomposites. Machine learning (ML) and X-ray Micro-Computed Tomography (micro-CT) are employed to bridge the gap between experimental measurements from DMA and computational modelling. A novel mesh-free computational approach, combining fast Fourier transform (FFT)-based homogenization and the Lippmann-Schwinger (LS) method, was applied to analyze the porous heterogeneous microstructures. The analysis of pore geometry and fiber distribution, along with the associated stress-strain behavior, provides valuable information regarding stress concentration at critical material interfaces. The proposed method revealed a higher Von Mises stress and strain in the matrix surrounding the fiber ends, a principal locus of load transmission. Further, the experimental DMA results highlight the importance of considering interfacial adhesion, friction, segmental mobility, and intercalation effects on modulus, T<inf>g</inf>, and tan ?. PETG +15 wt % SCF demonstrated high damping (tan ?: 0.19) and a 35 % and 122 % rise in modulus in glassy and rubbery states, respectively. Meanwhile the relative modulus of PETG +1 wt % OMMT-NC + 5 wt % SCF and PETG +3 wt % OMMT-NC + 5 wt % SCF nanocomposites improved by over 41 % in the glassy state. © 2025 Elsevier B.V.