Just another WordPress site - Ruhr-Universität Bochum

Abstract

Virtual creep tests using phase-field simulations are a promising source of microstructural evolution image data in lieu of the expensive interrupted creep tests. Firstly, image analysis techniques will be used to extract relevant microstructural features from images obtained from phase field simulations [1]. Subsequently, machine learning techniques will be used for microstructure-property analysis such as relationship between γ and γ’ morphology and strain. As a preliminary work to utilize such a data-driven approach, microstructure images of a Co-based superalloy specimens from creep test interrupted at various strain levels up to 2% strain at 950 ºC/200 Mpa were acquired. Image analysis techniques using OpenCV package [2] in Python programming language were implemented to extract quantitative microstructural features. Three features, that are thickness and total perimeter of γ phase channels, and volume fraction of γ’ phase, showed patterns and clustering with variation in strain. Subsequently, a k-Nearest Neighbors classifier was trained using these three features as X variables and creep strain value as the Y variable (class). The model showed a prediction accuracy of up to 75%; here prediction accuracy is defined as the percentage of correctly classified new microstructure images. Such classifier can be used to determine the creep strain value from the microstructure of a material subjected to creep. However, to utilize such a data-driven approach to its true potential, image data at smaller strain intervals is required. Therefore, the support of phase-field simulations as a source of data will be taken. References: [1] Muhammad Adil Ali, High temperature creep in Ni-based superalloys, Master Thesis, 2018, Interdisciplinary Centre for Advanced Materials Simulation, Ruhr-Universität Bochum. [2] https://docs.opencv.org/3.1.0/d6/d00/tutorial_py_root.html