Abstract
The development of eutectic compositionally complex alloys (ECCAs) has attracted significant interest over the years due to their excellent castability and superior mechanical properties derived from multi-phase architectures. However, the exploration of ECCAs remains challenging within their vast and intricate compositional space, largely owing to the lack of reliable phase diagrams for high-order systems. Conventional empirical approaches are often inefficient and costly, hindering the rapid discovery of novel eutectic compositions and impeding the advancement of ECCAs as structural materials. Furthermore, these alloys often suffer from the classical strength–ductility trade-off. Typical ECCAs exhibit sharp phase interfaces only nanometers wide. While such boundaries can hinder dislocation motion, they also serve as stress concentrators that may lead to premature failure.To address these issues, this dissertation employs machine learning (ML) to accelerate the design of ECCAs with enhanced mechanical performance and investigates their underlying deformation mechanisms. In Chapter 2, an interpretable ML framework integrating a conditional variational autoencoder (CVAE) and an artificial neural network (ANN) is developed to enable direct generation of promising ECCA compositions. To mitigate data imbalance—a common issue in data-driven alloy design—a tailored data preprocessing strategy was implemented, incorporating thermodynamics-informed descriptors and K-means clustering. This approach facilitated the successful identification of several previously unreported dual- and tri-phase ECCAs across quaternary to senary systems.
Chapter 3 focuses on an Al-Co-Cr-Fe-Ni ECCA identified via the ML framework. Through thermomechanical processing, its chemical and microstructural heterogeneity was precisely controlled, resulting in an exceptional strength–ductility balance. Microstructural analysis and phase-field simulations revealed that a deformation-induced FCC-to-BCC transformation contributes strengthening, while the engineered hybrid microstructure enhances toughness by promoting distributed microcracking and improving fracture resistance.
Additionally, an ML-aided strategy was introduced to design ECCAs with compositionally graded interfaces (GIs) in Chapter 4. Using candidates generated by the CVAE model, an ANN classifier—trained with physics-based features encoding chemical segregation tendency—was developed to predict interfacial segregation behavior. A Cr-based ECCA with high predicted segregation propensity was selected and synthesized. The as-cast alloy exhibited GIs approximately 20 ± 5 nm wide, arising from Cr segregation along phase boundaries as predicted. After thermomechanical processing and annealing at 800 °C, the average GI width increased significantly to 86 ± 47 nm due to enhanced Cr diffusion. The structurally graded interfaces led to outstanding mechanical properties: the alloy demonstrated a yield strength of 1087 MPa and a uniform elongation of 24.5% at room temperature, markedly surpassing conventional Cr-based alloys. Experimental characterization revealed that the GIs serve a dual role: they act as strong yet deformable barriers that accumulate dislocations and enhance strain hardening, while also permitting dislocation transmission across interfaces upon reaching a critical stress. And the MD simulation reveals that wide GIs serve to enhance deformation compatibility between the soft and hard phases by promoting dislocation activity in the BCC phase.
| Date of Award | 2 Jan 2026 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Yong YANG (Supervisor) |