Real-time monitoring dislocations, martensitic transformations and detwinning in stainless steel: Statistical analysis and machine learning

Abstract

Acoustic emission (AE) of 316L stainless steel with of low Ni content shows, under tension, simultaneously three avalanche processes. One avalanche process relates to the movement of dislocations, the others to martensitic transformations and detwinning/twinning. Detwinning/twinning occurs predominantly at the early stage of the plastic deformation while martensitic transformations only become observable after large plastic deformation. All processes coincide during deformation with variable effect on AE. An excellent fingerprint for the detection of the coincidence between the several mechanisms is the observation of multivalued E ~ A2 correlations. AE signals from moving dislocations decay more slowly (~ 7×10−3s) and show long avalanche durations. In contrast, AE signals during martensitic transformations and detwinning/twinning decay rapidly (<4×10−4s) and show short avalanche durations. They can be distinguished by different energy exponents of their avalanches. The energy distributions of the mechanisms differ because energies are defined as the integral over the squared AE amplitudes, where the integration extends over the avalanche durations. A combination of statistical analysis with Convolutional Neural Network calculations provides an accurate and straightforward method for online, non-destructive avalanche monitoring of strain-induced martensitic transformations in 316L steel under high strain.