Physical Metallurgy Guided Machine Learning for Strain-Induced Precipitation of Nb (C, N) Based on the Orthogonalized Small Data

Strain-induced precipitation (SIP) plays an important role in controlling the microstructure and properties of micro-alloyed steels during hot rolling. However, due to lack of systematic experiments, the existing precipitation data are scarce and insufficient for accurate modeling of SIP behavior with good extrapolation. Herein, the data space of Niobium Carbonitride (Nb(C,N)) SIP kinetics is analyzed based on principle of orthogonal design, and the missing points in the orthogonal space are identified and supplemented in addition to the available data. The parameters in models for precipitation start and finish times are optimized by using the precipitation-time-temperature curves through genetic algorithm, and their relationships with compositions and processing conditions are established by support vector machine. Based on the orthogonally modified dataset, the machine learning (ML) models outperform the classical and ML models with the original data in terms of accuracy and are in good agreement with theoretical expectations. By using the new ML modeling, the evolution behavior of Nb (C, N) during hot deformation is predicted and verified with transmission electron microscope, and the influences of Nb content in steel, strain, strain rate, and the sensitivity of compositions and processing conditions on precipitation kinetics are quantitatively analyzed.