The influence of volume fraction on the compressive behaviour of a polymeric syntactic foam: From manufacturing to machine learning-based modelling

This study presents a comprehensive investigation of an epoxy-based syntactic foam with microsphere volume fractions (VFs) ranging from 0 % to 60 %, encompassing manufacturing, experimental testing, and both data-driven and numerical modelling. The manufacturing approach was designed to improve microsphere dispersion, material homogeneity, and process reproducibility. By integrating degassing and mould rotation during curing, the approach significantly reduced void content and improved uniformity compared to conventional mix-and-cast techniques. Compression experiments were conducted across all VFs following the manufacturing process, revealing a significant influence of microsphere content on the compressive strength,which decreased from 102 MPa at 0 % VF to 27 MPa at 60 % VF. To model the mechanical behaviour of the manufactured foams, two complementary strategies were developed and evaluated: (1) machine learning (ML) models trained on experimental stress-strain data, and (2) a finite element (FE) model using a representative volume element (RVE) in Abaqus CAE. The best-case machine learning model achieved an overall mean stress prediction error of 5 MPa (4.16 %), outperforming the RVE model (6.88 MPa) and providing reliable predictions even in the post-densification regime. The ML models achieved highly accurate predictions of the full stress-strain response and demonstrated strong potential as efficient and reliable tools for reproducing the mechanical behaviour of syntactic foams. • Enhanced microsphere dispersion achi-eved through degassing and mould rotation. • Microsphere volume fraction significantly influences compressive strength. • Data-driven model accurately predicts post-yield behaviour and energy absorption. • Machine learning model outperforms finite element model in accuracy and efficiency.