Influence of Chemical and Structural Complexity on Material Performance by Atomistic Modeling and Data-Driven Analysis

Abstract

The pursuit of advanced materials with superior mechanical properties has been a long-standing objective due to their extensive applications in various fields. Recently, structurally and chemically complex materials have demonstrated remarkable mechanical performance compared to conventional materials. Multicomponent alloys with chemical complexity, such as high entropy alloys, can be engineered to achieve high yield strength without compromising ductility. Structurally disordered systems like metallic glasses exhibit exceptional mechanical strength, often surpassing that of their crystalline counterparts. These materials with structural complexity, characterized by the absence of long-range order, exhibit high elastic limits, making them ideal for demanding applications. Given their superior properties, understanding the underlying reasons why chemical and structural complexity leads to these advantages is crucial for material design. While the structure-property relationship in crystalline materials is well-defined and comprehensible, it becomes significantly more complicated in structurally and chemically complex materials due to their heterogeneous atomic arrangements and diverse chemical compositions. Established models for defects in crystalline materials, such as solid solution strengthening and grain boundary strengthening (Hall-Petch strengthening), highlight clear structure-property correlations. However, deciphering this relationship in complex materials is difficult due to their disordered local structures and chemical heterogeneity. Establishing a structure-property relationship is further complicated by the inherently non-uniform nature of these materials. For example, the solute itself may not be well-defined in heavily concentrated multi-component alloys, rendering the classical solid solution strengthening model inapplicable. And structurally complex materials can have no well-defined defects, making traditional strengthening mechanisms ineffective. In this dissertation, we aim to deepen our understanding of how chemical and structural heterogeneity influence the properties of materials. To achieve this, we combine molecular dynamics simulations with data-driven techniques to reveal new physics-based insights. To address these challenges, we present three key projects as examples to demonstrate our approach. The three main examples are shown as follows: (1) We study the thermal and kinetic properties at chemically heterogeneous interfaces and employ a physics-based machine learning approach to predict the energetics and kinetics of Cr atoms within multicomponent Fe-Ni-Cr alloys. By establishing predictive maps based on local electronegativity and free volume, we provide new insights into manipulating the thermal stability and mobility of Cr atoms, which could facilitate the design of materials with superior properties. (2) We also investigate the interfacial mechanical behavior under external loading. It is observed that there is nonmonotonic effect of chemical heterogeneity on interfacial crack growth at high-angle grain boundaries. Our findings reveal that the chemical composition can be tuned to minimize or maximize non-affine deformations and thus control crack propagation for different design purposes. By selecting appropriate elements in grain boundaries, one can optimize depending on the desired application. (3) We then focus on ZrCu-based metallic glasses, a structurally complex material, and demonstrate how machine learning can concurrently predict global energetics and indicate structural heterogeneity. The results highlight the potential of machine learning techniques to bridge the gap between experimental observations and theoretical models in the study of disordered materials. In summary, we employ a novel approach to study and understand the structure-property correlation in chemically and structurally complex materials. By combining atomistic modeling with physics-based data-driven techniques, we establish models that offer accurate predictions and novel insights into material behavior. Ultimately, this study enhances our ability to develop next-generation materials with improved performance for advanced technological applications.