In-situ Imaging for Temperature Estimation in Laser Directed Energy Deposition with Wire Feedstock Using a Convolutional Neural Network

Accurate temperature estimation is crucial in metal additive manufacturing ensuring component quality and process efficiency. This study introduces the use of Convolutional Neural Networks (CNNs), namely MobileNet and ResNet, to predict temperatures directly from melt pool images without the need for extensive preprocessing. Through comparative analysis, MobileNet demonstrated superior performance over ResNet, achieving a mean absolute error of 0.0562 and a correlation coefficient of 0.9900. These findings underscore the effectiveness of CNNs in real-time temperature prediction tasks within Laser-Directed Energy Deposition with wire (DED-LB/w), highlighting significant advancements and setting the stage for further technological enhancements.