검색 상세

Direct Prediction of Microstructure Evolution Utilizing Multi-Physics Informed Generative Artificial Intelligence

  • 이현주 (Hyunjoo Lee, 서강대학교 일반대학원)

초록 (요약문)

The evolution of microstructures in materials significantly impacts their properties and performance, particularly in energy storage systems like lithium (Li)-metal batteries. The performance of Li-metal batteries is critically hindered by the formation and growth of dendrites, which can lead to short circuits and capacity loss. This dissertation explores novel approaches for the direct prediction of dendrite growth using advanced generative artificial intelligence and physics-informed neural networks. The phase field model developed in this dissertation provides a robust framework for accurately depicting the complex physical phenomena involved in dendrite formation. This model not only captures the intricate details of the dendrite growth process but also serves as a critical foundation for training advanced generative artificial intelligence models using physics-governed data. However, traditional phase field modeling, while effective in capturing the complex kinetics of dendrite growth, is computationally intensive, particularly for large domain sizes. Generative Adversarial Networks (GANs) were initially employed to directly predict dendrite growth patterns based on surface roughness data. Although GANs demonstrated the ability to generate high-fidelity synthetic data, their lack of physical constraints limited their predictive accuracy. To address this, Physics-Informed Neural Networks (PINNs) were integrated into the GAN framework, resulting in Physics-Informed Generative Adversarial Networks (PIGANs). The PIGAN model, incorporating fundamental physical laws, enhanced the accuracy of dendrite growth predictions by ensuring that the generated data adhered to critical physical principles. By embedding multiple physical principles, including conservation laws and surface energy minimization, the Multiphysics-Informed Generative Adversarial Networks (Multi-PIGANs) model achieved predictions that closely matched the detailed outputs of phase field simulations while significantly reducing computational costs. This model also enabled large-scale simulations of dendrite growth, providing comprehensive insights into the effects of surface roughness and other factors on dendrite formation and evolution. The dissertation concludes that integrating multiple physical laws into generative models offers a transformative approach for the direct prediction of microstructural evolution in materials science. The Multi-PIGAN framework provides a practical and efficient tool for simulating dendrite growth on a large scale, crucial for optimizing the design and reliability of Li-metal batteries. These advancements pave the way for future research and applications in developing next-generation energy storage systems, highlighting the potential of combining machine learning with physical principles to address complex scientific and engineering challenges.

more

목차

I. Introduction 1
A. Degradation of Lithium (Li)-metal Batteries Due to Dendrite Growth 4
B. Advancements in Predicting Microstructure Evolution 6
II. Phase Field Modeling of Dendrite Growth in Li-metal Batteries 10
A. Governing Equations 10
Ⅱ.A.1 Dendrite Growth Model 10
Ⅱ.A.2 Dead Li Accumulation Model 14
B. Numerical Methods 18
C. Results and Discussion 22
III. Prediction of Dendrite Growth Utilizing Generative Artificial Intelligence (AI) 39
A. Generative Adversarial Networks (GANs) Model 41
B. Results and Discussion 49
IV. Multiphysics-informed Generative Adversarial Networks (Multi-PIGAN) 53
A. Physics-informed Neural Networks (PINNs) Model 53
B. Multi-PIGAN Model 56
C. Results and Discussion 59
V. Conclusion 66
Bibliography 69

more
반출 Meta View 목록