3D EDX-FIB map of a WC-reinforced Aluminide Iron based coating: multivariate analysis and deep learning hybrid approach

Serial sectioning tomography (SST) in the focused ion beam-scanning electron microscope (FIB-SEM) constitutes an extremely powerful tool to capture subsurface microstructure in three dimensions. When coupled to energy-dispersive spectral mapping (EDS), FIB-SEMs can generate analytical information of the sampled volume. However, despite day-long acquisition times, low signal-to-noise ratio forces to choose between spectral precision and spatial resolution. In addition, simple window integral algorithms are inappropriate to appreciate the complex and rich data generated by EDS-SST. On one hand, multivariate analyses can certainly tackle the data complexity but not the load. On the other hand, the low signal-to-noise level severely impedes the generation of an acceptable training dataset for deep learning. There is therefore an opportunity to leverage the capability of multivariate analyses algorithm to classify low SNR data to produce the training dataset for the deep learning step. Here, we test this idea using a multi-phasic tribological coating made of tungsten carbides in an iron aluminide matrix. The coating was deposited using high-velocity oxyfuel jet of a ball-milled powder containing 30 % vol WC in Fe3Al powder.