Machine learning for microstructures classification in functional materials

A.K. Choudhary, A. Jansche, Grubesa Tvrtko, T. Bernthaler, G. Schneider
Aalen University,

Keywords: machine learning, deep learning, kerr microscopy, materials characterization, quantitative microstructure analysis, permanent magnets


Quantitative microstructure analysis and controlling process parameters are vital for new material development and improving quality. Of these, material characterization is one of the major challenges faced in the field of materials research. The general approach is the assessment of quantitative properties, which are dependent on the use of destructive/non-destructive techniques. The consumption of NdFeB sintered permanent magnets has been increasing drastically with the increase in the number of electric vehicles (EV) and hybrid electric vehicles (HEV) [1]. The addition of rare earth (RE) metals to magnet improve its properties and has a direct impact on its costs. Therefore, there is a high demand for permanent magnets with less RE content or inexpensive RE metals [2]. The search for novel magnetic phases requires efficient quantitative microstructure analysis to extract information like phases, grain distribution, and domain patterns, and correlate it with its intrinsic magnetic parameters such as saturation polarization, anisotropy constant, Curie temperature, domain wall dynamics, etc. This helps in obtaining the optimized microstructures in magnets with good intrinsic magnetic properties. In this presentation, we use the classical machine learning approach and advanced deep learning algorithm for the extraction of microstructural information from the micrographs of NdFeB sintered permanent magnets (Kerr microscopy) as shown in the figure-1. Due to the complex microstructural features, it is not a feasible option to use traditional approaches of image analysis for extracting quantitative information from these micrographs. The performance of the trained models are compared to EBSD data and manually hand labeled dataset prepared by a subject expert. The model has proved to be efficient, robust to different magnet samples and time efficient method for grain and texture analysis in functional materials like magnets.