Data-Based Modeling of Electrochemical Energy and Thermal Systems: Fuel Cell and Lithium-Ion Battery

Abstract

As a solution to combat climate change and environmental pollution, electrochemical energy systems such as Proton Exchange Membrane Fuel Cell (PEMFC) and Lithium-Ion Battery (LIB) are being developed as the replacement for fossil fuel-powered combustion engines, especially for ground transportation and aviation applications. These electrochemical energy systems must be able to operate independently and in conjunction with each other by complementing their advantages and limitations, such as efficiency, range, thermal behavior, aging, and operating environment. This interoperability requires accurate real-time computational models to control, diagnose, and adapt according to field requirements. A typical electrochemical energy system model needs to incorporate effects related to reactant concentrations, system overpotentials, thermodynamics, porous media mechanics, membrane dynamics, gas diffusion, electrode degradation, electrolyte status, ion transport, and chemical kinetics across various operating conditions, all of which result in complex interactions affecting the accuracy and reliability of the system. Today, both PEMFC and LIB use complex computational physics-based fluid dynamics models in the product development phase, which requires enormous computational power and long lead times for iterative prototype improvements. On the other hand, both PEMFC and LIB rely on simple lookup tables and semi-empirical equations as plant models that require intensive calibration activity to determine the mode of control and diagnosis for automotive applications. However, considering the present-day automotive propulsion systems, which operate in widely varied applications and geographic locations and have short product development cycles, these approaches are not able to comprehend the complexities, hindering the ability of these systems to operate at their full potential and leading to catastrophic failures (e.g., Thermal runaway). Data-based modeling techniques are one of the potential solutions, which is quite in contrast with other empirical or physics-based models where the entire input-output relations of the model are established primarily based on the data. Data-based models use aspects of statistics, probability, and network architecture, avoiding the complexities of physics-based models and intensive calibration, providing better accuracy in most cases, primarily where the complex mechanisms can’t be modeled using specific governing equations, and fast, efficient computation with much less computational resource requirement.

This thesis focuses on data acquisition (identifying and collecting the relevant data) and data-based model development by incorporating machine learning algorithms and regressors to predict the system's performance, thermal behavior, aging, and faults in real-time (on-board diagnostics). Data for these models is acquired through two approaches: experimentation by utilizing Fuel Cell and Green Energy Lab facilities such as the Automated Battery Test Station (ABTS), G20 fuel cell automated test station, and by partnering with the relevant industry. In the second approach, data is generated by simulation of physics-based models (CFD, Semi-empirical, equivalent circuit models) that are experimentally validated in the literature and developed within the research groups of UWaterloo. Development of a data-based model includes the identification of feature vectors (inputs), prediction attributes (outputs), state estimates (internal parameters), non-linearity of the systems, correlation factors of various system entities, and application of machine learning techniques such as feed-forward artificial neural network, support vector machine classifier - regressor, along with their respective adaptations and calibration processes. The primary objectives of this study are to develop data-based models for three main application areas: (i) Prediction of PEMFC performance, internal states of the membrane, cell voltage degradation, and system outputs. (ii) Prediction of LIB heat release rate during discharge and thermal dynamics of an open system during an exothermic reaction. (iii) Prediction of fuel cell battery hybrid electric vehicle’s system dynamics and thermal behavior.

During this study, various data-based models were developed to tackle the problems encountered in fuel cell-battery hybrid systems, such as predicting the fuel cell performance, fuel cell voltage degradation, PEMFC membrane dynamics, lithium-ion battery thermal dynamics, thermal behavior during exothermic reactions and dynamics of fuel-cell battery hybrid system. The results presented in this study proved the data-based model’s applicability in surrogate modeling, real-time system monitoring, controls, and diagnostics of electrochemical energy systems both at the component level and system level. Additionally, the results implicate that the data-based model can serve as a complement and alternative to the traditional computational fluid dynamics models as well as complex physics-based and empirical models to predict thermal gradients and system internal states during multifaceted reactions.