Prediction of melt pool temperature in directed energy deposition using machine learning

Abstract

Predicting the melt pool temperature distribution and history in direct energy deposition (DED) is crucial for estimating the microstructure, porosity, and mechanical properties of DED-fabricated metal parts. While analytical and numerical modeling methods have been introduced to predict the melt pool temperature, the prediction accuracy of these methods is relatively low because the real-time melt pool temperature distribution is not considered. To address this issue, we developed a data-driven predictive model using machine learning to estimate the melt pool temperature during DED with high accuracy. Two machine learning algorithms, including extreme gradient boosting (XGBoost) and long short-term memory (LSTM), were used to build the predictive models. Experimental results have shown that both XGBoost and LSTM can predict the melt pool temperature with high accuracy. While XGBoost is more computationally efficient than LSTM, LSTM achieves higher prediction accuracy and better robustness.