Artificial intelligence techniques for thermomechanical property optimization of metal-PLA composites additive manufactured parts

Abstract

The control of thermomechanical behavior in additive manufacturing (AM) has become increasingly important, especially in industries requiring high dimensional stability. Among various AM methods, fused deposition modeling (FDM) is widely used for fabricating polymer-based composites due to its cost-effectiveness and versatility. However, optimizing thermal properties such as the coefficient of thermal expansion (CTE) in metal and fiber-reinforced composites remains a challenge. This study addresses the research question: How do different FDM process parameters affect the CTE of metal- and fiber-reinforced PLA composites, and how can artificial intelligence techniques be used to optimize these properties? To investigate this, three types of reinforced PLA composites-copper (Cu), bronze (Br), and carbon fiber (Cf)-were fabricated using FDM. A Taguchi L27 orthogonal array was employed to evaluate the effects of infill density (30-60-90 %), layer thickness (100-200-300 mu m), printing speed (40-80-120 mm/s), and raster angle (0 degrees-45 degrees-90 degrees). The CTE values were experimentally measured using thermomechanical analysis. Artificial neural network (ANN) and regression models were developed to predict CTE behavior, with the ANN further optimized using a genetic algorithm (GA). The results demonstrated that layer thickness had the most significant impact on CTE, followed by infill density and material type. The lowest CTE value (42.38 mu m/m-degrees C) was achieved using Cf/PLA composites under optimal conditions. The ANN-GA model outperformed the regression model in prediction accuracy, with R2 values of 98.38 % and 95.71 %, respectively, and maximum prediction errors below 1.17 %. This study contributes to the scientific understanding of thermal behavior in FDM-printed composites and provides a robust AI-based framework for optimizing CTE without extensive experimental trials. The findings offer practical implications to produce dimensionally stable components in high-precision applications.