Combining low-loss EELS experiments with machine learning-based algorithms to automate the phases separation imaging in industrial duplex stainless steels

Abstract

At present, transmission electron microscopy is regarded as the main option when dealing with phase characterization for materials at a nanometric scale. The development and improvement of complementary techniques such as energy-dispersive X-ray spectroscopy (EDS), electron energy loss spectroscopy (EELS), imaging detectors and associated computational methods provide a huge variety of choices to determine the composition and crystal structure at any region of a specimen. Despite all these advancements, phase identification for some specific materials having phases with similar structure or similar composition might still be a difficult procedure to perform. To overcome this difficulty, a new method is proposed which combines the acquisition of low-loss EELS spectra with a subsequent analysis driven by machine learning based algorithms (clustering). In the present research, this new approach is applied to a set of industrial duplex stainless steels having different compositions and it is contrasted to other characterization alternatives. As opposed to the other options, this method proved to be effective, reliable and time-saving.