Robust and time-efficient grain size analysis approach for Aluminum alloys using machine learning techniques
Amit Kumar Choudhary1,2, Andreas Jansche1, Alex Banholzer1, Timo Bernthaler1, Gerhard Schneider1
1Materials Research Institute, Aalen University, Aalen, Germany
2Karlsruhe Institute of Technology, Karlsruhe, Germany

For materials such as aluminium, which play a vital role in the advancement of scientific growth have wide applications. Understanding the relationships between composition, structure, processing parameters and properties helps in the development of improved materials. Materials scientists, chemists and physicists have researched these relationships for many years, until the recent past, by characterizing the bulk properties of materials and describing them with theoretical models.

Automated quantitative microstructure analysis is a widely accepted method for quality control in materials, especially for grain size analysis. The subjectivity of human experts leads to variability in microstructure classification when performed manually due to the presence of multiple phases and their complex underlying structures. Further, the appearance of the microstructure differs based on parameters such as heat treatment, etching condition, alloying elements, etc. Therefore, a robust and accurate classification of these microstructures is very important while hard to achieve.
The ability of machine learning models to overcome the challenge of developing complex multivariate functions by considering human expert knowledge has been one of the major factors in their success. Within the scope of this research, a grain analysis approach has been developed and evaluated for five different aluminium alloys (light microscopic images) using machine learning and computer vision techniques. Further, the results were evaluated as per the standards (ASTM E 112 -13) to evaluate its accuracy with respect to the ground truth (human expert). The performance measure in terms of the grain boundary detection, robustness and time efficiency was a major part of the experiments.