Comparison of Image Analysis Techniques for the Determination of the Influence of Melt Flow on the Solidification Microstructure of Technical Aluminum Alloys

Abstract

This work aims to contribute to the improvement of automated analysis of 2D microstructures in technical aluminum alloys by evaluating the effectiveness of various microstructural image analysis methods: Traditional analysis techniques for 2D images and cutting-edge computer vision deep learning-based methods (Mask Region-based Convolutional Neural Networks or Mask R-CNN [1], a model used to detect the position and exact shape of objects) are compared for detecting, classifying, and quantifying microstructural features in technical iron containing aluminum alloys. Determining that deep learning-based methods lead to comparable or better results would allow for such models to be used for analysis of the chosen microstructural features. This could improve detection accuracy and speed, as well as measurement reproducibility.

The focus is on detecting and characterizing Fe-containing intermetallic phases that precipitate during solidification, with the aim of examining the effects of increased iron content on the microstructure of these alloys. Therefore, the microstructure of directionally solidified hypoeutectic AlSi6Cu4Fe1 and hypereutectic AlSi6Cu4Fe2 alloys is analyzed with regards to area fraction and size of the intermetallic precipitates within two distinct sample processing zones: the first half of the samples was solidified under diffusive mass transport conditions and the second under forced convective conditions, both in microgravity. This dual approach allows for the evaluation of the impact of convection on the microstructure of aluminum alloys with elevated iron content.