Abstract Material Extrusion for metals (MEX/M) is an emerging multi-step and user-friendly Additive Manufacturing (AM) technology for the realisation of metallic parts. Key part properties, specifically surface roughness, density, and dimensional accuracy, are heavily influenced by the printing parameters. Optimising these parameters is fundamental to ensuring part integrity during the subsequent debinding and sintering (D&S) stages, which ultimately determine final performance and industrial viability. To address this challenge, the present work proposes a data-driven modelling framework based on a multi-model ensemble integrating different machine learning regressors. The ensemble surrogate model provides accurate and robust predictions of key quality metrics, enabling process optimisation through a response surface derived from explainable artificial intelligence (XAI) techniques. In particular, the final ensemble achieved leave-one-out cross-validation errors of 0.07 ± 0.06 g/cm³ for green density, 1.73 ± 1.49% for total dimensional deviation, and 2.20 ± 1.88 μm for lateral surface roughness, remaining within the experimentally defined process-variability thresholds. The optimised configuration improved green density by approximately 21% and reduced surface roughness by approximately 46%, demonstrating that the proposed XAI-assisted ensemble surrogate model can reduce experimental effort while supporting traceable and interpretable process optimisation. This approach not only identifies the optimal process parameters but also reveals the relative importance and physical influence of each parameter on the outcomes. The proposed method, demonstrated through experimental and validation procedures on highly-filled copper filament, significantly reduces the need for extensive experimental trials while supporting traceable, explainable, and certifiable process optimisation.
Pellegrini et al. (Sat,) studied this question.