In addition to prototyping, Powder Bed Fusion (PBF) AM processes have lately been more widely used to manufacture end-use parts. These changes lead to necessity of higher requirements to quality of a final product. Optimization of process parameters is one of the ways to achieve desired quality of a part. Finite Element Method (FEM) and machine learning techniques are applied to evaluate and optimize AM process parameters. While FEM requires specific information, Machine Learning is based on big amounts of data. This paper provides a conceptual framework on combination of mathematical modelling and Machine Learning to avoid these issue