Study on material behaviours of additively manufactured high-impact polystyrene using artificial neural networks

Abstract

Fused Filament Fabrication (FFF), a process parameters-dependent manufacturing method, currently dominates the additive manufacturing (AM) sector because of its prominent ability to produce parts with intricate profiles, customise products, and minimise waste. Though the effects of FFF process parameters were investigated experimentally, recent research highlighted the importance of developing numerical modelling and computational methods on optimising the FFF printing process and FFF-printed materials. This study aims to investigate the tensile strength (TS) of FFF-printed high-impact polystyrene (HIPS) via devising a systematic testing and analysis framework, which combines experimental testing, representative volume element (RVE)-finite element method (FEM), rule of mixture (ROM), and artificial neural networks (ANN). HIPS samples are fabricated using FFF considering the variations of infill density, layer thickness, nozzle temperature, raster angle, and build orientation, and tested with standard tensile testing. The rule of mixtures (ROM) and its modified version (MROM) are employed to calculate the TS of longitudinally and transversely built samples at various infill densities, respectively, while an ANN model is constructed to investigate the effect of material anisotropy precisely. The optimal ANN architecture is built with five hidden layers with the number of neurons in each layer as 44, 82, 169, 362, and 50. Although both MROM and ANN perform well on the validation set, ANN exhibits superior accuracy with only a maximum error of 0.13% for training set and 11% for validation set. The combination of the RVE-FEM, MROM, and ANN approaches can significantly improve the FFF printing process of polymers for optimisation.