September 5, 2026 — ebiotrade reports on the paper 'Towards real-time prediction of thermal history and hardness in laser powder bed fusion using deep learning,' published in Computational Materials Science. The study proposes a physics-informed deep learning surrogate model that enables real-time prediction of thermal history and hardness in laser powder bed fusion (LPBF). The model employs an LSTM sequence-to-sequence network trained on high-fidelity multi-scale finite element simulation data (Abaqus with Goldak double-ellipsoidal heat source and JMAK phase transformation model). Inputs include physical derivative features such as scan coordinates, laser power, scan speed, and layer thickness. By integrating thermal history with tempering coupling effects, the model performs point-by-point autoregressive prediction of temperature sequences for quenched and tempered steel AISI 4140 / 42CrMo4, achieving a test-set temperature error of approximately 2.6 ± 1.2 K (RMSE). Local Vickers hardness is then derived using the Hollomon–Jaffe relationship, enabling an integrated process-structure-property prediction framework. For powder buyers, the value is twofold: (1) Real-time thermal history prediction can guide customers in fine-tuning process windows by grade and batch, strengthening procurement loyalty toward traceable, batch-consistent powders; (2) Co-building a 'powder characteristics–thermal history–hardness' database with powder manufacturers offers a high-barrier collaborative entry point for serving tool steel and aerospace structural component customers.