Real-time layer height estimation during multi-layer directed energy deposition using domain adaptive neural networks

Abstract

Metal additive manufacturing (AM), such as laser direct energy deposition (DED), is gaining popularity because of its capability in manufacturing near-net-shaped complex components for various industrial applications. However, the geometry control during the DED process, especially at corners with sharp turns, remains a daunting task. To achieve geometry control, geometry estimation to identify the relationship between the process parameters and geometry attributes is vital. In this study, a real-time layer height estimation technique is developed for DED using a laser line scanner, vision camera, and domain adaptive neural networks (DaNN). An emphasis is placed on layer height estimation at sharp corners during multi-layer deposition. First, multi-layer straight-line deposition data is collected using laser line scanner and an initial layer height estimation model is constructed. Then, to efficiently achieve layer height estimation during corner deposition, an DaNN model is established using the multi-layer straight-line deposition data and the constructed initial model. The actual traverse speed at the corners is measured using a vision camera and fed into the DaNN model as one of input features. Finally, the DaNN model is updated online to further improve estimation accuracy during corner deposition. The proposed technique has been validated by DED experiments and the results show that the layer height can be estimated in 0.018 s with an average accuracy of 25.7 mu m when multiple layers with an average height of 250 mu m are deposited at corners with different angles.