Optimising surface roughness, dimensional accuracy and printing time of FDM PETG parts using statistical methods and artificial neural network

<ns3:p>The purpose of the paper is to optimise key quality characteristics of FDM PETG parts—surface roughness, dimensional accuracy, and printing time—by investigating the effects of critical process parameters. It further seeks to develop and compare predictive models using both statistical methods and artificial neural networks (ANN) to enhance the process efficiency and applicability of FDM technology.An experimental framework based on a Taguchi L25 orthogonal array was employed to analyse parameter combinations systematically. PETG specimens were modelled in SolidWorks, converted to G-code, printed, and evaluated. Statistical techniques, including ANOVA and regression analysis, were applied for parameter optimisation. Additionally, ANN models were developed to predict the output responses. The dual-methodology approach ensures robust modelling and validation of the results.The results demonstrate that both regression and ANN models effectively predict the responses of FDM PETG parts with high accuracy. However, the ANN models significantly outperformed the statistical methods, achieving superior predictive accuracy, particularly for surface roughness (R = 0.9944). Multi-objective optimisation identified optimal parameter settings that minimise surface roughness, dimensional errors, and printing time simultaneously.The study is limited to PETG material, and the findings may not be directly generalisable to other thermoplastics. Future research could include other parameters and mechanical properties that define the overall bearing capacity of the material, as well as investigate the effects of post-processing techniques.The research offers practical insights for manufacturing high-quality PETG parts with enhanced surface finish and dimensional accuracy, achieved in a reduced printing time. The developed models offer valuable guidance for optimising FDM processes in industrial applications.The research presents a novel integration of statistical methods and artificial neural networks for the multi-objective optimisation of FDM PETG parts. The study demonstrates the superior predictive capability of ANN modelling in capturing complex parameter-response relationships, while providing validated optimal processing parameters that simultaneously enhance surface quality, dimensional accuracy, and manufacturing efficiency. The work establishes a comprehensive framework for intelligent process optimisation in industrial additive manufacturing applications.</ns3:p>