Predictive modeling of additively manufactured carbon fiber-PLA mechanical components via ML

Abstract

Purpose The study aims to predict and optimize two critical parameters, surface roughness and energy consumption, in additive manufacturing (AM) processes using carbon fiber-reinforced polylactic acid (PLA) material. These parameters are essential for enhancing the efficiency and quality of AM-produced components. Design/methodology/approach A mechanical connector was fabricated using the AM process, employing the Box-Behnken experimental design method with four input parameters: layer thickness (LT) (150-200-300 mu m), infill density (ID) (40%-80%-100%), nozzle temperature (NT) (200-210-220 degrees C) and printing speed (PS) (40-80-120 mm/s). Predictive models were developed using four machine learning (ML) algorithms: Gaussian process regression (GPR), extreme gradient boosting (XGBoost), artificial neural network (ANN) and random forest regression (RFR). Model performance was evaluated using mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE) and R-squared (R-2). Additionally, ANOVA was conducted to identify the most influential parameters on surface roughness and energy consumption. Findings The RFR model demonstrated superior accuracy with low error values and high R-2 scores in estimating both surface roughness and energy consumption. ANOVA results indicated that LT (43.96%) and PS (40.01%) were the most significant factors affecting surface roughness, while LT (50.45%) and ID (27.58%) significantly influenced energy consumption. Originality/value This study underscores the effectiveness of ML algorithms and statistical analysis in modeling and optimizing AM processes. The findings provide valuable insights into improving the efficiency and quality of 3D-printed components, particularly through the integration of carbon fiber-reinforced materials and advanced predictive modeling techniques.