Bayesian approach for inferrable machine learning models of process-structure-property linkages in complex concentrated alloys

The difference in the mechanical behaviors of dilute solid solutions, complex solid solutions and their corresponding strengthening mechanisms, is an evolving field of study. An understanding of the mechanisms and formulation of theories of strengthening in the complex atomic energy landscapes could eventually lead to a better understanding of the fundamental behavior of condensed matter itself. In this work we attempt to extract the effect of thermo-mechanical processing on the microstructure-mechanical property linkages of complex concentrated alloys (CCAs) by training machine learning (ML) models using processing information / parameters as features. The effect of processing on the phase morphology and the mechanical properties is studied. The stacking fault energy (SFE) predicted based on CCA composition is used as a benchmark to identify deformation mechanisms that are activated based on the arrangement of the component elements within the distorted CCA lattice. This work presents a novel method that attempts to establish ML based process-structure-property (PSP) linkages that could help capture higher order dependencies that may not be adequately captured by existing relations between mechanical properties, phase evolution, composition and processing information. An assortment of Bayesian-learning models are used to create a framework that captures the evolution of phases, their volume fractions, grain sizes and the corresponding change in mechanical properties of a diverse set of CCA compositions as they encounter various processing conditions. The evolution of the mechanical property with grain size is captured as Hall-Petch relations as an example of possible PSP linkage representations.