OPTIMIZATION OF FDM PROCESS PARAMETERS FOR TENSILE STRENGTH USING MACHINE LEARNING AND EVOLUTIONARY ALGORITHMS

In this study, the authors examined a data-driven approach for enhancing the tensile properties of ABS components manufactured via fused deposition modeling (FDM). Test specimens were printed using systematically varied combinations of extrusion temperature, layer thickness, and infill density. The resulting tensile data were then used to train several regression models. Among the approaches examined, gradient boosting, support vector regression, and polynomial regression proved to be the most dependable, each accounting for more than eighty-six percent of the variation in measured strength. These trained models were then paired with evolutionary optimization methods to search for printing conditions that would yield higher-performing parts. Multiple optimizers converged on parameter regions associated with predicted tensile strengths in the upper 30-megapascal range. To provide a physical reference point for these predictions, a finite element model of the ISO-527-5A specimen was built in COMSOL. The simulated tensile response reproduced the expected stress distribution in the gauge section and indicated a failure load consistent with both the machine learning estimates and the manufacturer’s reported properties. While the optimized parameter sets could not be fabricated within the project timeline, the close agreement between simulations, model predictions, and known material behavior supports the reliability of the proposed workflow. The results highlight the usefulness of integrating supervised learning, optimization, and finite-element analysis to guide FDM process decisions and demonstrate the potential for extending this approach to applications requiring more complex mechanical or functional performance criteria.