Application of an RBF Neural Network for FDM Parts' Surface Roughness Prediction for Enhancing Surface Quality

To improve the surface roughness of parts fabricated using fused deposition modeling, modeling of the surface roughness distribution is used before the fabrication process to enable more precise planning of the additive manufacturing process. In this paper, a new methodology based on radial basis function neural networks (RBFNNs) is proposed for estimation of the surface roughness based on experimental results. The effective variables of the RBFNN are optimized using the imperialist competitive algorithm (ICA). The RBFNN-ICA model outperforms considerably comparing to the RBFNN model. A specific test part capable of evaluating the surface roughness distribution for varied surface build angles is built. To demonstrate the advantage of the recommended model, a performance comparison of the most well-known analytical models is carried out. The results of the evaluation confirm the capability of more fitted responses in the proposed modeling. The RBFNN and RBFNN-ICA models have mean absolute percentage error of 7.11% and 3.64%, respectively, and mean squared error of 7.48 and 2.27, respectively. The robustness of the network is studied based on the RBFNN's effective variables evaluation and sensitivity analysis assessment for the contribution of input parameters. Finally, the comprehensive validity assessments confirm improved results using the recommended modeling.