Machine learning image-based analysis for bead geometry prediction in fused granulate fabrication for large format additive manufacturing.

Abstract

This study investigates an artificial neural network (ANN) for predicting cross-sectional geometry in fused granulate fabrication (FGF), a key polymer-based technology in large-format additive manufacturing (LFAM). Critical process parameters-layer height, transverse speed, and screw speed-were systematically varied to study their effects on bead morphology. A full factorial design generated a robust training dataset, and cross-sectional images were processed for model training. The ANN architecture, featuring two hidden layers, was paired with image processing techniques to manage computational demands. Results showed strong agreement between predicted and experimental cross-sections, with a mean absolute error of 8.88%, highlighting the ANN's capability in capturing geometry. This approach advances prior LFAM studies by predicting full cross-sectional images rather than contour points, improving complex shape prediction. The findings demonstrate the ANN's effectiveness for FGF profiles and its potential to enhance geometric precision and generate complex shapes across LFAM technologies.