| Popis: |
Global warming is currently considered one of the most crucial challenges worldwide. The necessity of reducing greenhouse gas pollution requires the development and utilization of renewable technologies capable of storing and delivering clean energy. Rechargeable batteries represent a prominent solution to achieve this goal. However, various limitations still prevent these devices from a wide-spread application substituting the existing polluting energy resources. The improvement of rechargeable batteries could be achieved, for example, with the appropriate engineering of electrochemical, thermal, and mechanical processes occurring at the atomic, nano, and micro scales, while incorporating accurate monitoring of a battery’s functionality integrated into dynamic systems at the macro scale. Metallic and high entropy alloy (HEA) nanoparticles (NPs) are of great interest in various lithium-based batteries like lithium-oxygen batteries (LOBs) due to their superior electrocatalytic properties. Thus, improving battery’s constituent materials at the nanoscale is important because the understanding at the fundamental level yields better predictive capability, which allows optimal design and operation of new battery systems. The scope of my thesis is to contribute to the investigation of features of rechargeable batteries at different scales. At the nanoscale, metallic and HEA NPs have been investigated, while at the macroscale the state-of-charge (SOC) and state-of-health (SOH) in different types of battery electric vehicles (BEVs) have been studied. The recent advancement of artificial intelligence (AI) techniques, based on machine learning (ML) and deep learning (DL) algorithms, have been used in this research to perform different types of predictions on simulated data generated through xvii multi-physics simulations, using experimentally acquired data for a final physical validation. At the nanoscale, fully convolutional neural networks (FCNs), trained on simulated transmission electron microscopy (TEM) images, like high-resolution transmission electron microscopy (HRTEM) and scanning transmission electron microscopy (STEM) images, have been used to estimate the 3D structure of gold (Au) and quinary HEA NPs, by predicting the number of atoms in their atomic columns, a feature known as column heights (CHs). The realistic 3D configuration predicted for these NPs in experimental images indicate a satisfactory performance of the developed model when applied to real-life cases. At the macroscale, the forecasting of the SOC has been performed with learning algorithms for time-series estimation applied to simulated data including the time series of electrical and mechanical variables describing the dynamics of the Tesla S and Nissan Leaf BEVs. The obtained results suggest that training a learning algorithm on variables describing the entire dynamic of the vehicles leads to a more precise SOC estimation compared to the state-of-the-art methods including only the battery’s electrical variables. A study of the SOH decay caused by the formation of the solid electrolyte interface (SEI) has also been performed. The results emerged from the study of the 3D NPs reconstruction, allow quantitative structure property relationship (QSPR) analysis, for the engineering of rechargeable batteries materials, aimed to improve representative macroscale parameters of the batteries like SOC and SOH. In fact, a potential outcome of this work is to combine the results of the nanoscale study of NPs into the macroscale investigation of the SOC and SOH, which could provide a more effective characterization and design of rechargeable batteries. Despite the primary objective of investigating rechargeable batteries at different scales, this work also establishes a basis for combining AI algorithms with modeling data, obtained with multi-physics and multi-scale simulations, which could have a broader impact in mechanical engineering applications. |
