Surface roughness prediction in fused deposition modelling by neural networks

Fused deposition modelling is a proven technology for the fabrication of both aesthetic and functional prototypes. The obtainable roughness is the most limiting aspect for its application. The prediction of surface quality is an essential tool for the diffusion of this technology, in fact at product development stage, it allows to comply with design specifications and in process planning it is useful to determine manufacturing strategies. The existing models are not robust enough in predicting roughness parameters for all deposition angles, in particular for near horizontal walls. The aim of this work is to determine roughness parameters models reliable over the entire part surface. This purpose is pursued using a feed-forward neural network to fit experimental data. By the utilisation of an evaluation function, the best solution has been found. This has been obtained using a feed-forward neural network for fitting the experimental data. The best solution has been founded by using an evaluation function that we constructed. The validation proved the robustness of the model found to process parameters’ variation and its applicability to different FDM machines and materials.