| Author: | Wang, Wei |
| Title: | Phase-field fracture modeling and numerical solver acceleration |
| Advisors: | Ruan, Haihui (ME) |
| Degree: | Ph.D. |
| Year: | 2026 |
| Department: | Department of Mechanical Engineering |
| Pages: | xxvi, 163 pages : color illustrations |
| Language: | English |
| Abstract: | In this thesis, we develop two phase-field fracture models (PFFMs) to investigate: (1) the mechanical failure mechanisms in all-solid-state batteries (ASSBs) using Landau-Ginzburg framework, and (2) the fracture behavior of ion-exchange (IOX)-strengthened cover glass using a variational framework. Although the phase-field (PF) method mitigates the nonlinearity of crack propagation by introducing a diffusive interface and is compatible with finite element (FE) solver, the standalone FE fracture simulations remain computationally intensive. To overcome this limitation, we propose a numerical solver acceleration, particularly through AI-accelerated FE framework, to enhance computational efficiency. This approach offers a promising route for efficiently solving highly nonlinear PFFMs in large scales and in multi-physical fields, eventually achieving quantitative analysis to further guide experiments. ASSBs are anticipated to achieve exceptional energy density and enhanced safety, enabled by the direct use of lithium metal anodes and the suppression of dendrites through solid-state electrolytes (SSEs). However, recent experimental studies have revealed discharging-induced void formation at lithium/SSE (Li/SSE) interfaces and charging-induced cracks within SSEs, both of which facilitate lithium penetration, finally resulting in mechanical failure of cell. To address these challenges, we develop a high-fidelity mechano-electrochemical model for ASSBs, particularly focusing on the coupled kinetics of void and crack evolution. This model demonstrates that void growth results from lithium stripping, driven by the disparate interfacial and bulk diffusivity of lithium. This process results in nonuniform Li⁺ distribution during electroplating, subsequently localized interfacial damage zone, SSE cracking, and ultimately, lithium plating within cracks. Our findings indicate that while high stack pressures reduce void size, they do not prevent cracking; in contrast, optimized lateral compressive stresses can effectively suppress SSE fracture and inhibit dendrite formation. Ion-exchange (IOX)-strengthened cover glass has become ubiquitous in portable electronics due to its exceptional fracture resistance. However, establishing a quantitative relationship between residual stress profiles and fracture strength remains elusive, as the experimental data are highly scattered due to the uncertainties in surface flaws severity. To address this challenge, we develop a variational PFFM to systematically investigate the fracture mechanisms and the stress-shielding effects of surface compressive stress (CS) on pre-existing surface flaws. Our study specifically evaluates the strengthening effectiveness of different combinations of surface CS and depth of layer (DOL) under compression, employing both ring-on-ring (ROR) and ball-drop (BD) test. Simulations reveal that surface CS plays a more dominant role in crack inhibition for ROR test; while for BD scenarios, DOL emerges as the more critical parameter. These insights are illustrated through comprehensive fracture strength contour maps and BD height analyses. Numerical solvers for partial differential equations (PDEs) often face a trade-off between computational efficiency and accuracy. Neural operators (NOs) offer a promising avenue to accelerate simulations, but they are hindered by several challenges: the need for large training datasets, error accumulation in dynamic settings, and poor generalization in multi-physics scenarios. To tackle these problems, a novel hybrid framework is proposed, which integrates physics-informed deep operator network (PI-DeepONet) with finite element method (FEM) via domain decomposition. The core innovation lies in efficient coupling FEM and DeepONet subdomains using a Schwarz alternating method, expecting to solve complex and nonlinear regions by a pre-trained DeepONet, while the remainder is handled by FEM. The Newmark-Beta time-stepping scheme is embedded directly into the DeepONet architecture, substantially reducing long-term error propagation. Furthermore, an adaptive subdomain evolution strategy enables the machine learning (ML)-resolved region to expand dynamically, capturing emerging fine-scale features. The framework's efficacy has been rigorously validated across a range of solid mechanics problems—spanning static, quasi-static, and dynamic regimes for either linear elasticity or hyper-elasticity—demonstrating accelerated convergence rates (up to 20% improvement over traditional FE coupling approaches) while preserving solution fidelity with error margins consistently below 3%. The validation in mechanics shows the potential of this framework to further achieve high-fidelity multi-physical simulations. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14358