Experimental Study and ANN Development for Modeling Tensile and Surface Quality of Fiber-Reinforced Nylon Composites

Abstract

Additive manufacturing (AM) is gaining widespread adoption in the manufacturing industry due to its capability to fabricate intricate and high-performance components. In parallel, the increasing emphasis on functional materials in AM has highlighted the critical need for accurate prediction of the mechanical behavior of composite systems. This study experimentally investigates the tensile strength and surface quality of carbon fiber-reinforced nylon composites (PA12-CF) fabricated via fused deposition modeling (FDM) and models their behavior using artificial neural networks (ANNs). A Taguchi L27 orthogonal array was employed to design experiments involving five critical printing parameters: layer thickness (100, 200, and 300 mu m), infill pattern (gyroid, honeycomb, and triangles), nozzle temperature (250, 270, and 290 degrees C), printing speed (50, 100, and 150 mm/s), and infill density (30, 60, and 90%). An analysis of variance (ANOVA) revealed that the infill density had the most significant influence on the resulting tensile strength, contributing 53.47% of the variation, with the strength increasing substantially at higher densities. In contrast, the layer thickness was the dominant factor in determining surface roughness, accounting for 53.84% of the variation, with thinner layers yielding smoother surfaces. Mechanistically, a higher infill density enhances the internal structural integrity of the parts, leading to an improved load-bearing capacity, while thinner layers improve the interlayer adhesion and surface finish. The highest tensile strength achieved was 69.65 MPa, and the lowest surface roughness recorded was 9.18 mu m. An ANN model was developed to predict both the tensile strength and surface roughness based on the input parameters. Its performance was compared with that of three other machine learning (ML) algorithms: support vector regression (SVR), random forest regression (RFR), and XGBoost. The ANN model exhibited superior predictive accuracy, with a coefficient of determination (R-2 > 0.9912) and a mean validation error below 0.41% for both outputs. These findings demonstrate the effectiveness of ANNs in modeling the complex relationships between FDM parameters and composite properties and highlight the significant potential of integrating ML and statistical analysis to optimize the design and manufacturing of high-performance AM fiber-reinforced composites.