A comparative study of artificial neural network with pattern search algorithm and response surface methodology for multi-objective optimization of dimensional accuracy and mechanical properties in PLA-based fused filament fabrication

Abstract

In additive manufacturing, the optimisation of process parameters to simultaneously enhance both mechanical and geometrical properties remains a significant challenge. This study presents a comparative analysis between two distinct approaches to multi-response optimisation: a hybrid method combining Artificial Neural Networks with Pattern Search Algorithm, and the traditional Response Surface Methodology. Both methods were applied to optimise the process parameters of Fused Filament Fabrication using Polylactic Acid. The hybrid ANN + PSA model was used to predict process outputs and optimise printing parameters, while RSM employed a statistical approach to identify optimal parameter combinations through designed experiments. Results show that ANN + PSA achieved better outcomes with a maximum tensile strength of 61.88 MPa and dimensional deviations within 3%, compared to RSM's tensile strength of 53.05 MPa and deviations up to 4%. The ANN + PSA model demonstrated higher predictive accuracy with an R² of 94.34% during training and 91.34% during evaluation, versus RSM’s R² of 90.5% for tensile strength prediction. Additionally, ANN + PSA consistently required fewer experimental trials due to its integration with the Pattern Search Algorithm, improving computational efficiency. The findings demonstrate that ANN + PSA is computationally efficient for scenarios with fewer experimental trials, while RSM offers a more detailed understanding of interaction effects. The comparative insights from this study contribute to enhanced multi-response optimisation strategies in additive manufacturing.