| 總結: | <p>This thesis focuses on the study of electrochemical reaction and mass transport processes via numerical simulation, parameter estimation and deep learning enabled inferences. The significance of computation to understanding of electrochemistry will be shown and discussed.</p>
<p>Chapter 1 presents the fundamentals of electrochemistry, including concepts, theories and experimental practices to provide background information to this thesis. Chapter 2 provides the basics of simulation, including simulation of mass transport and models for interfacial reactions and boundary conditions. Testing and validation methods are also introduced.</p>
<p>Chapter 3 analyzes the Tafel slope of cyclic voltammetry recorded for a heterogeneous electrochemically reversible electron transfer process coupled with non-linear homogeneous chemical kinetics. Tafel plots are shown to be markedly non-linear with potential, indicating the essential need to be analysed via simulation rather than approximate analytical theory. A super-Nernstian Tafel slope is observed for EC2 reaction. Chapter 4 models a non-unity stoichiometric reversible electrode reaction, using the oxidation of bromide with coupled chemical kinetics as an example.</p>
<p>Chapter 5 generalizes the simulations of the previous chapters with a focus on the effect of coupled preceding or following chemical reaction on Tafel analysis predicted with finite difference simulation. Four cases of coupled chemical reactions are elaborated.</p>
<p>Chapter 6 introduces the application of artificial intelligence (AI) in electrochemical studies, using neural networks to extract rate and equilibrium constants from voltammograms of dissociative CE reaction and predict voltammograms from these constants. AI provides an excellent tool to map the relationship between experimental data and parameters of interest. The training data of neural networks is facilitated by numerical simulation, an optimal alternative to experimental data which is hard and expensive to collect and process. Chapter 7 introduces the “simulation, experiment and machine learning” framework for parameter estimation and is validated with the dissociation of acetic acid experiment: by training neural network with simulated steady state current to correlate with kinetic and thermodynamic constants of acetic acid dissociation, AI can accurately predict these constants based on experimental data. This framework provides a new tool for parameter estimation by combining simulation and AI for extracting data from experiments.</p>
<p>Chapter 8 introduces the state-of-art physics-informed neural network (PINN) for solving partial differential equations (PDEs). By imposing physical constraints in the form of PDEs on neural network, the PDEs can be solved without discretization. PINN is applied to solve diffusive mass transport problems in electrochemical reactions with boundary conditions enforcing interfacial reactions to predict voltammetry. PINNs are applied to a variety of 1D and 2D models, introducing a faster and/or easier way of simulation of voltammetry.</p>
<p>Chapter 9 elaborates on further exploration of PINN notably in hydrodynamic voltammetry. Using curriculum learning, PINN can solve the convective-diffusion problem of a single microband channel electrode and predict the mass transport limited current. PINN can also predict the collection-efficiency of double microband channel electrode in generator-detector formation. Furthermore, convective-diffusion coupled with chemical kinetics is also solved to predict the kinetically controlled mass transport limited current. The results are validated by experimental data and/or analytical expression where applicable.</p>
|
|---|