Novel method for predicting the cracks of oxide scales during high temperature oxidation of metals and alloys by using machine learning

Abstract

In high temperature processes, the degradation of the materials is the major problem that may lead to higher maintenance costs and accidents. One form of the failure is the crack and spallation of the protective oxide film due to either the mechanical stress developing during the oxidation process or the thermal stress resulting from the mismatch of thermal expansions of oxide formed and alloy. Conventionally, the Pilling-Bedworth ratio (PBR) was used for predicting the crack and spallation of oxide by determining the volume changes of oxide and alloy. Although, the PBR is simple but it gives the inaccurate predictions. Therefore, in this work, the machine learning was used instead of PBR for predicting the formed of oxide produced and its spallation in the temperature range of 600–1200 ^oC. The alloy compositions, the oxide formed during oxidation, oxidation conditions and periods were inserted in the model as the inputs. From the results, the random forest with 15 estimators was the best machine learning model providing more than twice accuracy for predicting oxide spallation than using PBR.