| Abstract Scope |
In this work, a combined discrete element method-computational fluid dynamic (DEM-CFD) modeling and deep neural network (DNN) machine learning (ML) algorithm was used to predict microstructural feature and defect formation, such as grain size, grain morphology, porosity, and lack of fusion, etc. in additively manufactured IN718 alloy. The process variables (power, speed, layer thickness, etc.) and mechanistic variables (thermal gradient, cooling rate, etc.) were used as inputs to the ML model. To generate data set for ML algorithm, an experimental, numerical model and cellular automaton (CA) results were used. The inputs and outputs of ML model are selected based on a novel governing physics approach, which then proposed. The DNN model quickly predicts the results and the predictions are in good agreement with the numerical as well as experimental results. This work presents an avenue for future physics-based optimization to control the microstructure evolution and defect formation in additive manufacturing. |