Surrogate modeling of melt pool temperature field using deep learning

Powder-based additive manufacturing has transformed the manufacturing industry over the last decade. In the Laser Powder Bed Fusion (L-PBF) process, a specific part is built in an iterative manner in which two-dimensional cross-sections are formed on top of each other by selectively melting and fusing the desired areas of the powder bed. In this process, the temperature field and melt pool morphology play a major role in the quality of the manufactured part and its possible defects. Therefore, predicting these factors is of high importance. However, simulating such a complex phenomenon is usually very time-consuming and requires huge computational resources. In this work, we create three datasets consisting of single-trail L-PBF processes using the Flow-3D simulation software and use them to train a convolutional neural network capable of predicting the three-dimensional temperature field solely by taking the process parameters and the time step as input. The CNN achieves a relative Root Mean Squared Error less of than 5% for the temperature field in the solidifying region and an average Intersection over Union score of 80% to 90% in predicting the three-dimensional geometry of the melt pool. Moreover, since time is included as one of the inputs of the model, the temperature field can be obtained in a matter of a few seconds for any arbitrary time step without the need to iterate and compute all the steps.