A data-driven ANN model for estimation of melt-pool characteristics in SLM process

Abstract

Abstract: The Selective Laser Melting (SLM) process is layered to manufacture a near-net-shaped metallic 3D component. The SLM process involves multiple physical phenomena during part fabrication. A highly intense moving laser heat source causes rapid temperature rise leading to melting and flow of raw material in the melt-pool and subsequent solidification during the process. The present study develops an analytical model by considering various physical phenomena of the SLM process, such as heat transfer, fluid flow, the Marangoni effect, etc. It is solved using an in-house developed analytical formulation based upon Finite Volume Method (FVM) approach. Further, a data-driven machine-learning based on an Artificial Neural network (ANN) model with feedforward neural architecture was trained to estimate the melt-pool dimensions for the single-track SLM process. The developed ANN model was tested for Ti6Al4V alloy by considering all the thermo-physical material properties. A total of 1224 datasets were generated using the analytical formulation to train, validate, and test the neural network. The 70–80% of the training, validation, and testing data fall between the deviation range of 23.55 μm and −21.96 μm. The model predicts melt-pool characteristics with an average error of 0.085%, 0.053%, and 0.056% for depth, width, and length, respectively.