Application of Neural Network to Predict Printability of Polycaprolactone Using FDM

Abstract

Fused Deposition Modelling (FDM) has gained its extensive application because of its lower cost, simple machinery and ease in operation. The simplicity of construction has enabled the development and use of several opensource FDM 3D printers. However, the FDM process parameters are difficult to tune as they significantly affect the printability of the material used. Many studies have been made to obtain a relationship between these parameters and print quality. The high non-linearity due to the numerous process variables playing their role complicates this relationship. Artificial neural networks can handle a high number of process variables and aid in studying the effect of these process variables on the fabricated part. This work aims to use a neural network for predicting whether a given set of parameters can lead to a successfully fabricated Polycaprolactone (PCL) part using FDM. A two-layer feed-forward neural network has been developed to predict the printability of PCL based on a given set of process parameters. Various combinations of process parameters were used to fabricate a PCL part using an opensource FDM 3D printer, and a data of 45 such combinations were used to develop the neural network. Three different neural network models with 10, 20 and 30 number of hidden neurons were trained. Finally, a group of process parameters was fed to the network with the best performance, and the network provided the probability of the given group of process parameters to result in the successful fabrication of the PCL part as the output.