Mechanical properties of metals and their alloys are strongly governed by their microstructures. Therefore, image analysis as a method of microstructural investigation is essential to gain valuable knowledge on processing – microstructure – property relationships.

In the presented work, nanometer-sized precipitates of a hardenable wrought aluminum alloy and their microstructural changes due to aging at elevated temperatures are studied using dark-field transmission electron microscopy (DF-TEM) images. The precipitates act as obstacles to dislocation motion within the material and are therefore critical to the mechanical performance of the components.

We propose a quantitative analysis technique for precipitates based on a deep learning approach using a small dataset. With this technique, we introduce a training scheme to overcome the challenges of detecting precipitates in heterogenous raw DF-TEM image data. This approach would significantly improve the segmentation models and facilitate automatic detection and extraction of precipitates and quantification of their relevant dimensions.

The work presented is part of Materials-open-Laboratory (Mat-o-Lab)