Convolutional LSTM based melt-pool prediction from images of laser tool path strategy in laser powder bed fusion for additive manufacturing

As the importance of mass customization production is increasing, additive manufacturing have been an emerging method of end-product process not only prototype production. Despite continuous improvements, metal processes, such as Laser Powder Bed Fusion, face challenges such as inconsistent output quality and the need for extensive post-processing. Recent approaches are the real-time adjustments of laser power along the meltpool size captured by an online camera. This method, however, still consume the cost and time to find optimized process plan. This paper introduces a novel approach that predicts melt-pool size and thermal peaks based on tool path strategies, without the trial-and-error of actual processes, by using pre-trained deep learning models. Convolutional Long Short-Term Memory algorithm-based model converts enhanced images of toolpath data-including laser coordinates and laser paramenters-into the size graph of meltpools. Using an open dataset of 12 different cubes made by additive process, The deep learning model is trained with tool path strategies and meltpool images in advance. The proposed model detects intensive meltpool peaks to choose optimal laser powers during bi-directional and spiral tool paths.