4-5. Machine learning approach for the prediction of microstructure-property relationships in Ti alloys

Gopal Viswanathan¹, Santosh Koduri², Peter Collins³, Hamish Fraser¹

1. The Ohio State University, Columbus OH 43210, USA; 2. Intel Corporation, Hillsboro, OR-97124, USA; 3. Iowa State University, Ames, IA-50011, USA

Abstract: The prediction of microstructure property relationships based on physically relevant phenomenological model in complex metallic system is not realistically possible. An alternative approach based on neural network offers a solution to this problem. In this talk an application of neural network for the prediction of microstructure/property relationships in Ti alloys is presented. Such an approach, however, invariably requires sufficiently large data bases covering accurate descriptions of input microstructures to predict truthful properties. This requires effective way of producing high-fidelity descriptions of microstructures without compromising the physical principles that govern the material behavior. The first part of the talk will focus on our experience in producing and testing such microstructures in α/β Ti alloys with particular focus on the predictive capability for tensile properties. The second part of the talk will include our experience with materials characterization tools and techniques such as aberration corrected scanning transmission electron microscope (STEM), atom probe and X-Ray tomography, electron back-scattered diffraction (EBSD)and electron channeling contrast (ECCI) to discern the underlying mechanisms which in turn help improve the predictive capabilities.