Integrating data-driven system to predict temperature and distortion in multi-layer direct metal deposition processes

This study proposed a framework to train an artificial neural network (ANN) by a data-driven system to predict the temperature and distortion in multi-layer direct metal deposition (DMD) of SS 304. By integrating thermomechanical variables, the research ensures the fidelity of finite element (FE) simulations, which are validated against existing data. Notably, the study achieves enhanced precision over prior work by varying the heat input sources and heat transfer equations. A novel aspect of this research is the use verified FE simulation to add data to data-driven system to train an efficient ANN for predicting temperature and distortion based on key parameters such as laser power and scanning speed. The iterative process involved multiple FE simulations with varying laser parameters to refine the ANN's predictive capabilities. This methodology enabled the identification of relationships between manufacturing parameters, temperature, and distortion. The iterative training continued until the ANN's predictions and subsequent FE simulation results converged within an acceptable margin. The findings confirm that the trained ANN can predict temperature and distortion both accurately and expediently, marking a significant advancement in the control of the DMD process.