Deep Learning Based Prediction of Thermal History During The Laser-Based Powder Bed Fusion (PBF-LB) Additive Manufacturing Process

Laser-Based Powder Bed Fusion (PBF-LB) is a key technology in additive manufacturing (AM) for metals, offering a high degree of design freedom for complex geometries. Accurate modeling of the thermal history and resulting material properties is critical, especially for steels with solid-state phase transformations. This research integrates high-fidelity multi-scale finite element method (FEM) simulations and long-short term memory recurrent neural networks (RNNLSTM) to efficiently predict thermal histories. The RNN-LSTM model, trained on FEM-generated data, demonstrated high prediction accuracy for demonstrators of different sizes. The performance of the model was trained and evaluated for three datasets, with the steady-state dataset yielding the highest prediction accuracy. This approach can improve material characterization in PBF-LB, optimize printing strategies, and elucidate the relationship between process parameters and the resulting microstructure.