Optimising 3D-printed carbon fibre composites using machine learning: Balancing strength and efficiency
José Humberto S. Almeida, Guilherme Ferreira Gomes
Abstract
Additive manufacturing (AM) of fibre-reinforced composites offers design freedom but necessitates a trade-off between mechanical performance and production speed. This study introduces a data-driven framework that integrates machine learning (ML) and genetic algorithms (GA) to optimise interlaminar strength and minimise printing time simultaneously. Among the seven ML models evaluated, artificial neural networks (ANNs) achieved the highest predictive accuracy (9.2% for strength, 14.7% for time). Pareto-optimal solutions were identified, with the best configuration reaching 27.6 MPa SBS in 75.9 minutes. Experimental validation confirmed the predicted failure modes, including interlaminar shear and delamination. Additionally, computed tomography (CT) scans revealed distinct internal microstructures: the strength-optimised configuration exhibited a well-consolidated morphology, while the speed-optimised sample showed increased void content. These results demonstrate that the proposed optimisation framework not only enhances AM efficiency and mechanical performance, but also enables more predictable failure behaviour, enabled by informed microstructural control. • Machine learning improves interlaminar strength and reduces printing time. • Artificial Neural Network outperforms other surrogate models. • Optimised parameters enhance manufacturing time and mechanical performance. • Data-driven framework ensures scalable, reliable 3D printing of composites. • Surrogate models enable strong and fast-printed composites.