Materials Informatics-Assisted Optimisation and Random Forest Prediction of Mechanical Properties in FDM-Printed PLA/MWCNT Nanocomposites

Abstract

Additive manufacturing has become one of the mainstream manufacturing technologies for producing polymer components with complex geometries and high material utilisation. Among various additive manufacturing technologies, fused deposition modelling (FDM) is widely applied due to its strong adaptability and low cost. However, the mechanical properties of polylactic acid (PLA) components printed via FDM are highly dependent on printing parameters and the reinforcing materials used. This study will carry out a systematic investigation centred on these relevant influencing factors. This study uses PLA composite filaments containing 0.4 wt.%, 0.8 wt.% and 1.2 wt.% multi-walled carbon nanotubes as reinforcing materials. Three categories of process parameters are controlled, nozzle temperature (206°C, 210°C, and 214°C), infill density (30%, 60%, and 90%), layer thickness (0.10, 0.14, and 0.18 mm) and experiments are designed using gyroid infill and the Taguchi L9 orthogonal array. Tensile testing complies with ASTM D638, and impact test was conducted with ASTM D6110 standards. Signal-to-noise (S/N) ratio analysis and their further investigation through analysis of variance (ANOVA) were used to analyse the experimental results to determine the effect of each process parameter and to find out optimal printing conditions. The main contributions of all factors to both tensile strength and impact strength were,infill density had the biggest contribution, followed by layer thickness and nozzle temperature. The highly recommended parameter combination results in a maximum tensile strength of 43 MPa and an impact strength of 3.664 kJ/m² at the infill density of 90%, layer thickness of 0.18 mm, and nozzle temperature of 214 °C. Scanning electron microscopy (SEM) fracture morphology characterization was conducted on PLA-multi-walled carbon nanotube (MWCNT) composite samples. For the optimised samples, MWCNTs are uniformly distributed, and interlayer bonding strength is improved; the non-optimised samples have defects including microvoids and poor interfacial bonding, which limit energy dissipation. A self-built random forest regression model can verify experimental results, predict the mechanical properties under the optimal printing parameters, and deliver reliable prediction accuracy.