Abstract
Surface roughness is one of the major challenges in parts produced by Fused Filament Fabrication (FFF) because of the staircase effect created during layer-by-layer fabrication. This study focuses on minimizing the surface roughness of FFF-printed nylon composites reinforced with 20% carbon fiber using a combination of Taguchi design, Multi-Criteria Decision-Making (MCDM) techniques, and Machine Learning (ML) algorithms. Five process parameters, namely build orientation, layer thickness, infill density, raster angle, and infill pattern, were varied at different levels using a Taguchi L18 experimental design. Surface roughness parameters, namely arithmetical mean height of the roughness profile (Ra) and maximum height of the roughness profile (Rz), were measured in two perpendicular directions using a MarSurf VD280 surface profilometer. The experimental results were first analyzed using the Signal-to-Noise (S/N) ratio method. Signal-to-noise analysis identified layer thickness as the most influential factor within the investigated parameter space, while MANOVA indicated a significant multivariate association of layer thickness with the combined surface roughness responses under the fitted main-effects model (Wilks’ Lambda, p = 0.033). Multi-response optimization was then carried out using a comparative Entropy-TOPSIS and PCA–TOPSIS approaches to identify the optimal process settings for improved surface quality. The resulting multi-response performance index was further used to train and evaluate several regression-based machine learning algorithms with different train-test split ratios (60:40, 70:30, 80:20, and 90:10). Among the algorithms tested, the AdaBoost regression model showed the best predictive performance with a maximum coefficient of determination (R2) value of 0.99. The analysis revealed that layer thickness was the most influential parameter affecting surface roughness. Surface characterization using Scanning Electron Microscopy (SEM) confirmed the presence of voids and surface irregularities in the printed specimens. The proposed hybrid optimization and prediction framework demonstrates an effective approach for improving the surface quality of carbon fiber reinforced nylon components manufactured through FFF.
Keywords: Entropy; principal component analysis; TOPSIS; Adaboost algorithm; signal to noise ratio