Modeling an Accurate ANN Model with Multiple Inputs to Predict Dimensional Accuracy

Conventional models such as regression models are unable to fully capture the underlying relationship between the inputs and outputs of complicated processes such as fused deposition modeling (FDM). This study aims to develop an artificial neural network (ANN) model, a non-conventional model better suited to communicating the relationship between for most influential FDM process parameters such as layer height, infill density, printing speed and printing temperature to the single response of dimensional accuracy. 78 samples were generated using face centered central composite design (FCCCD), printed using the Ender 3 V2 printer and measured for errors using Vernier calipers. Utilizing the data, several networks were trained with varying hidden layers and hidden layer neurons to identify the best-performing models. The most accurate models were developed and compared based on the lowest mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and highest coefficient of determination (R2). Results deduced that the best-performing ANN model structure is 4–12–12–1 with the lowest overall MSE of 0.002898 and highest overall R2 of 0.955769.