X80 pipeline steel is widely used in oil and gas pipelines because of its excellent strength, toughness, and corrosion resistance. It is welded via gas metal arc welding (GMAW), risking high cold crack sensitivities. There is a certain relationship between the joint hardness and cold crack sensitivity of welded joints; thus, predicting the joint hardness is necessary. Considering the inefficiency of welding experiments and the complexity of welding parameters, we designed a set of processes from temperature field analysis to microstructure prediction and finally hardness prediction. Firstly, we calculated the thermal cycle curve during welding through multi-layer welding numerical simulation using the finite element method (FEM). Afterwards, BP neural networks were used to predict the cooling rates in the temperature interval that ferrite nuclears and grows. Introducing the cooling rates to the Leblond function, the ferrite fraction of the joint was given. Based on the predicted ferrite fraction, mapping relationships between joint hardness and the joint ferrite fraction were built using BP neural networks. The results shows that the error during phase fraction prediction is less than 8%, and during joint hardness prediction, it is less than 5%.
Read full abstract