Fig. 1. X-ray diffraction pattern for low-Ni 316L stainless steel before tension (a) and after tension (tensile strain ~70%) showing the additional martensite α’ peaks (b).

Fig. 2. (a) Stress-strain curve with the AE jerk spectra for low Ni-316L stainless steel sample under tension with a rate of 1×10-2mm/min. AE signals of dislocation movements (b) and martensitic transformations/detwinning-twinning (blue/green) (c). (d) Energy versus amplitude correlations for dislocations (red) and martensites/detwinning-twinning (blue/green) show an exponent x = 2.0. (e) Amplitude versus duration correlations for dislocations (red) and martensites/detwinning-twinning (blue/green). The exponent χ is around 1.5. (f) Energy versus duration correlations for dislocations (red) and martensites/detwinning-twinning (blue/green) show exponent γ close to 3.0.

Fig. 3. Electron Back-Scattered Diffraction for different deformation strain in 316L stainless steel with twins marked in red and martensite marked in blue. (a) Initial sample with 15% twins, (b) 6.6% twins, 0.05% martensites and dislocation slip (identified by the trace line) at 10.2% strain, (c) 6.69% twins, 1.84% martensites and obvious dislocation slip at 25.9% strain, (d) 4.9% twins and 8% martensites at 43.6% strain, (e) 3.8% twins, 51.2% martensites at 67% strain, (f) evolution of twin and martensite during tensile deformation.

Fig. 4. (a) Maximum likelihood (ML) curves with an energy exponent around 1.5 for dislocations, 1.8 for martensitic transformations/detwinning-twinning during the entire deformation process. (b) ML curves for martensitic transformations/detwinning-twinning occur at different strains. The exponents are above 2 for detwinning/twinning for strains <8%, and 1.8 for martensitic transformations for strains >8%. The exponent suggests a weak power law mixing when the strain is below 10%. (c) Probability density function (PDF) of dislocation movement in red and martensitic transformation in blue.

Fig. 5. Probability density function of AE duration for dislocation movements and martensitic transformations with exponents α = 2.5 and 2.3 for the two populations (a); typical profiles for dislocations (b) and martensitic transformations (c) with amplitudes around 130μV, and (d) typical profiles of detwinning/twinning.

Fig. 6. (a) Time sequence of normalized amplitude square of waveforms. The dislocation signals decay slowly with no difference between the elastic and plastic strain regimes. The martensitic transformations show very short lifetimes with a very rapid decay of the AE signals. (b) Histogram of time delays for two AE sensors detecting time difference of the same acoustic signals.

Fig. 7. Waiting time distribution of dislocation movements, martensitic transformations, and their correlations. The mean field power-law decay with 1-φ ~1 is seen only for dislocations. All other correlations follow approximately a Poisson distribution for inter event times greater than 1s.

Fig. 8. The schematics of the identification procedure based on CNN transfer learning and statistical analysis.

Fig. 9. Online monitoring module with the combination of CNN model and statistical analysis for a new data set. (a) As an example, eight waveforms were chosen from those collected by the AE equipment during the course of the experiment. (b) Time dependence of the avalanche energies with predictive label for the dataset. (c) Energy versus maximum amplitude correlation for the same signals as shown in (b). (d) Comparison between the online monitoring module and post-mortem statistical analysis. The blue, green and red dots represent martensitic transformations, detwinning/twinning and dislocation movements predicted by the online monitoring module. The open symbols represent martensitic transformations, detwinning/twinning and dislocation movements by statistical analysis after the measurement. The agreement between the two data sets is excellent.