Characterizing minor phases in engineering alloys with averaging and dictionary indexing of EBSD patterns

Session

PSA.2 - Metals & Alloys

Authors

Mr Håkon Wiik Ånes (1), Mr Vetle Rundestad Østerhus (1), Professor Jarle Hjelen (1), Professor Antonius Theodorus Johannes van Helvoort (2), Professor Knut Marthinsen (1)

Affiliations

1. Department of Materials Science and Engineering, Norwegian University of Science and Technology
2. Department of Physics, Norwegian University of Science and Technology

Keywords

aluminium, dictionary indexing, EBSD, minor phases, pattern averaging

Abstract text

Characterizing minor phases in microstructures is important in metallurgical alloy and process development. With electron backscatter diffraction (EBSD) in the scanning electron microscope (SEM), it is relatively easy to determine the crystal orientations of the major phases, e.g. of aluminium or ferrite and austenite in super-duplex stainless steels (SDSS). However, it can be challenging to determine the crystal structure and orientation of minor phases like Sigma and Chi in SDSS and particles containing Fe, Si and Mn in AA3xxx series Al alloys. This can be because the EBSD pattern (EBSP) shows overlapping Kikuchi bands as the electron beam-sample interaction volume contains both the major and minor phases. Or, the backscatter electron (BSE) yield on the EBSD detector is too low to determine any crystal structure or orientation. The minor phase’s crystal structure might also be unknown a priori, in which case all EBSPs must be tested against many candidate structures to find the best match, which can be time consuming.

To improve the identification of minor phases and their orientations, we explore masking a scan of EBSPs after the location and orientation of the major phases have been obtained. Thereby, the patterns from the minor phases can be studied in more detail. Upon masking, similar patterns belonging to e.g. a particle in Al can be averaged to increase the signal-to-noise ratio, effectively increasing the possibility to determine its crystal structure and orientation with relation to the matrix. Dictionary indexing is done with the open-source software EMsoft [1], where every experimental pattern is compared to a dictionary of dynamically simulated patterns. This has proven to be more robust towards pattern noise compared to traditional Hough indexing [2].

The approach is demonstrated here on a direct-chill cast, homogenized, as-deformed 95% cold-rolled Al-0.53Fe-0.15Si-0.39Mn (wt.%) alloy. EBSPs were acquired from a vibration polished surface in the ND-RD section using a NORDIF UF-1100 detector. We will also demonstrate this approach on identification of Sigma and Chi phases in SDSS [3].

The upper row in Figure 1 shows (a) a BSE image and Al (b) orientation similarity (OS) and (c) orientation maps obtained after dictionary indexing of (96 x 96) px EBSPs from the AlMn alloy. Note that they are cut-outs from a larger image and scan. The (d) bottom row shows four experimental patterns, of which the first three from the left belong to three particles highlighted in both the BSE image and OS map. The rightmost pattern is from an Al grain in between two of the particles. The OS per phase is derived from comparing the orientations of the top matching simulated patterns in a scan location to the orientations in the top matches in the four adjacent locations. High values indicate that they are similar. Low Al OS values in the shown map in most cases indicates either a grain boundary or a minor phase, in this case a particle. Thus, patterns from particles can be obtained by thresholding the OS map and masking out the Al patterns. Potentially, other information, like BSE yield on the EBSD detector, can be used to refine the thresholding. Patterns from a single particle can be clustered based on spatial proximity and subsequently averaged to obtain one pattern per particle with an improved signal-to-noise ratio. These reduced number of patterns can be compared to simulated patterns from multiple candidate phases. Here, they have been compared to candidate phases of Al₆Mn, Al₆Fe, α-AlMnSi, and α-AlFeSi, with the best matching simulated pattern and phase shown below each experimental pattern. Applying this to all detected particles can yield phase fractions, with some uncertainty based upon how well a particle matches the candidate phases. [4]

Figure 1: (a) Backscatter electron image and EBSD (b) orientation similarity (OS) and (c) orientation maps obtained by dictionary indexing from the same region of interest of a 95% cold-rolled AlMn alloy. Note that they are cut-outs from a larger image and scan. In addition, (d) experimental patterns and their best matching dynamically simulated patterns are shown. The three first patterns from the left are obtained by thresholding the OS map, clustering the patterns into particles, and then averaging these to increase their signal-to-noise ratio. The pattern on the right is from the Al matrix, which was indexed by comparing all experimental patterns to a dictionary of simulated Al patterns.

References

[1] YH Chen, SU Park, D Wei, G Newstadt, MA Jackson, JP Simmons, M De Graef, AO Hero, Microscopy and Microanalysis 21(3), 2015,  p. 739–752.

[2] HW Ånes, J Hjelen, BE Sørensen, ATJ van Helvoort, K Marthinsen, IOP Conference Series: Materials Science and Engineering (in review).

[3] VR Østerhus, HW Ånes, M Haukali, OM Akselsen, I Westermann, M Karlsen, J Hjelen, (poster submitted to EMC 2020).

[4] The authors would like to thank Hydro Aluminium Sunndalsøra for providing the material. HWÅ acknowledges NTNU for financial support through the NTNU Aluminium Product Innovation Centre (NAPIC). The authors also thank staff at the Electron Microscopy Lab at NTNU for maintaining the microscopy facilities.