Abstract: Recognition microstructure of alloys is crucial for materials design and development. In this work, a web-based tool was established for automated microstructure recognition and mechanical properties prediction of polycrystalline alloys. Firstly, combining deep learning models with image-processing algorithm, a AI modulus was developed to recognize the characteristics of grain boundary (GB) in polycrystal-structured images, associated with statistical analysis of grain size, distribution and grading. After training over 500 images of polycrystalline Mg alloys, GB identification can be achieved with an accuracy of over 85%. Then, the GB recognition results can be imported into a microstructural-based ffnite element (FE) model as the geometry input to further mechanical simulation. Afterward, using the local stress/strain contour results predicted from FE models as training set, another deep learning model was built to accelerate mechanical prediction of alloys. Using corresponding Mg alloy as an example, the model achieves an accuracy over 82%, while reducing computation time from hours to seconds. This work aims to provide tools and data support for rapidly establishing structure-property relationships in polycrystalline alloys. The developed web-based tool is currently available for free trial at http://120.46.222.99:5000/.

Keywords: Polycrystalline; microstructure recognition; mechanical properties; machine learning