Artificial neural network-based construction of a processing map for P550 steel for non-magnetic drill collars

A processing map was developed for P550 steel using data-driven modeling. Flow stress data were obtained from hot compression tests conducted at temperatures of 900-1200 °C and strain rates of 0.01-10 s -1 . The predictive performance of the Arrhenius, artificial neural network (ANN), and gradient boosting regression (GBR) models was compared. The ANN model exhibited the highest accuracy, with an R 2 of 0.998, Average Absolute Relative Error of 3.08%, and Root Mean Square Error of 4.6 MPa. In contrast, the Arrhenius and GBR models showed lower accuracy, with R 2 values of 0.954 and 0.995, respectively. Accordingly, the ANN model was employed to generate the flow stress data used for processing map construction. The optimal processing window was identified in the range of 950-1200 °C and 0.01-1 s -1 . Microstructural analysis under representative deformation conditions confirmed increasing dynamic recrystallization and decreasing deformation microstructures with rising temperature. Instability regions were also experimentally validated, further supporting the reliability of the ANN-based processing map.