Impact of feature selection on neural network prediction of fused deposition modelling (FDM) print part properties

Fused deposition modelling (FDM) is a popular additive manufacturing technique due to its low cost of producing complex parts. The quality of print part from the FDM process in terms of energy demand, mechanical and physical properties are influenced by its process parameters. The task of using artificial intelligence to better understand the influence of process parameters on the quality characteristics of FDM manufactured parts is constantly being explored. This study, on the other hand, aimed to implement feature selection methods to determine the optimal process parameters for accurately predicting print part quality characteristics. The particle swarm optimization (PSO) and neighbourhood component analysis (NCA) methods were used to select only relevant process parameters that provide a significant contribution to the development of the artificial neural network (ANN) model. The results showed that the NCA-ANN model is the best predictor of energy consumption, ultimate tensile strength, part weight and print time. Furthermore, the features from PSO contributed to PSO-ANN being the best average hardness predictor. It can therefore be established that incorporating the feature selection technique of PSO and NCA to elect only important process parameters can improve the prediction performance of the FDM print part property ANN model.