Towards generalizable thermal profile prediction in additive manufacturing using constrained random walks and machine learning

Defect formation in additively manufactured materials is strongly governed by the thermal history experienced during printing, which depends on process parameters that interact in complex and often non-linear ways. A finite element method (FEM) model was developed to extract thermal profiles at representative locations in laser powder bed fusion (L-PBF) processes, iterating over a defined set of printing parameters under a single scanning strategy. Among the machine learning architectures evaluated, Neural Networks (NN) provided the most accurate and robust predictions. To generalise the approach beyond fixed scanning strategies, a constrained random walks (CRW) methodology has been introduced to generate training datasets with sufficient variability. The resulting model predicts thermal histories accurately and at a fraction of the computational cost of FEM, providing a practical basis for process optimisation and defect control in L-PBF.