Machine learning-enhanced 3D analysis of tantalum-containing particles impact on creep performance in creep strength enhanced ferritic steels

Abstract

While the tempered martensitic microstructure of Creep Strength Enhanced Ferritic (CSEF) steels provides remarkable creep resistance, long-term stability remains a challenge. Microstructural risk factors in CSEF steels caused by nitrides, inclusions and coarsening of particles compromise mechanical integrity, accelerating crack initiation and propagation. Tantalum (Ta) additions promote the formation of nanoscale MX carbonitrides that hinder dislocation recovery and stabilise the microstructure, extending creep life. However, the unintended formation and coarsening of large Ta-containing particles could potentially counteract these benefits by serving as preferential sites for creep cavity nucleation. To better understand the direct correlation between creep damage and coarse Ta-containing particles, 3D Focused Ion Beam (FIB) serial sectioning, Scanning Electron Microscopy (SEM), Transmission Electron Microscopy (TEM) and machine learning (ML) enhanced image segmentation was performed on a novel CSEF steel. High-resolution SEM imaging of sequential 100 nm thick slices enabled 3D reconstruction and quantification of particle distributions. ML enhanced image segmentation significantly accelerated the phase separation process across the large dataset, enabling efficient and consistent analysis that would be impractical with manual methods. The results reveal a close spatial association between the coarse Ta-containing particles and creep damage, with progressive accumulation of these particles toward the fracture surface (volume percentage increasing from 0.038% in as-received material to 0.093% near creep fracture), suggesting these particles serve as preferred sites for cavity nucleation with substantial impact on creep behaviour. This study paves the way for microstructural risk factor assessment in next-generation high-temperature materials and highlights the advantages of combining FIB-SEM serial sectioning with ML to analyse complex microstructures.