Speaker
Description
Thermomechanical processing (TMP) of high-strength low-alloy (HSLA) steels relies on the careful control of the interaction between recrystallization and strain-induced precipitation during austenite conditioning. A comprehensive understanding of this interaction requires the simultaneous quantification of prior austenite grain size evolution and niobium carbonitride precipitation across a wide range of deformation conditions. These tasks remain experimentally challenging due to the statistical significance required and the multiple length scales involved.
In this work, we propose a characterization methodology combining experimental and AI-based methods for the systematic study of the correlation between grain size, recrystallization, and strain-induced precipitation in a Nb-bearing HSLA steel (0.04 wt.% Nb). Double-hit compression tests were performed in a Gleeble simulator at 975 °C with strains ranging from 0.15 to 0.4 and interpass times from 5 to 4000 s, covering the full range of recrystallization kinetics including the characteristic stasis plateaus.
The methodology integrates three complementary approaches: (i) a deep-learning-based semantic segmentation model for the automated reconstruction of prior austenite grain boundaries on light optical micrographs, enabling statistically significant grain size distributions over large areas; (ii) a machine-learning model for the segmentation of Nb(C,N) precipitates in STEM images on carbon extraction replicas, overcoming the limitations of conventional thresholding in the presence of microstructural relief; and (iii) matrix dissolution combined with ICP-OES analysis, used to normalize the volume fraction derived from extraction replicas and to correct the overestimation inherent to Ashby's equation. A filtering strategy is additionally proposed to exclude copper sulfides from the precipitation quantification.
This approach reveals a continuous evolution of grain size and precipitate size distribution during recrystallization stasis, challenging the assumption of constant grain size in existing models and establishing a basis for the systematic study of microstructure evolution during TMP.