Abstract
Lithium-ion batteries (LiBs) are widely used energy storage resources and provide a supply for many electrical applications, such as electric vehicles (EVs), and grids, etc. This is due to LiBs characteristics such as lightweight, high-energy density, compact size, and extended life. Despite their advantages, LiB performance is inherently subject to decreases over time due to the degradation of their electrochemical components. Moreover, in service, they may fail to supply the required energy, leading to the breakdown of the whole host application. Therefore, monitoring battery degradation and accurately predicting the remaining useful life (RUL) in LiBs has become critical to enhance performance as well as optimising battery life. It is imperative to consider various parameters and conditions, including but not limited to the consistent operating temperature, charging and discharging cycles, voltage levels, current usage patterns, state of health (SoH), and prevailing environmental factors such as humidity and pressure. Furthermore, the calibration and validation of the prediction model must align with the specific type of LiBs and their applications, employing historical data and usage statistics to optimise accuracy and reliability in predicting RUL, thereby enhancing both performance and optimisation of battery life. For this reason, this thesis focuses on accurately estimating the RUL of LiBs by developing a framework of a model-based battery coupled with a novel adaptive filter technique.The thesis presents a novel approach for predicting the RUL of LiBs using the Smooth Particle Filter (SPF) Based Likelihood Approximations algorithm. The proposed algorithm provides various advantages over the classic Particle filter method, including the capability to handle complicated nonlinearities and uncertainties in battery behaviour. Hence, the majority of methods necessitate special consideration to enhance the estimation's convergence rate and stability. The intrinsically noisy estimation is the fundamental obstacle in predicting the likelihood functions and derivatives. Consequently, there is an urgent need for a comprehensive and adaptable battery model, coupled with a precise nonlinear estimation algorithm, to address these challenges and improve battery RUL prediction. The results showed that the proposed algorithm can predict the RUL of LiBs with high accuracy and convergence rate compared to the classic PF algorithm, making it a promising tool for battery management systems.
Moreover, an improvement to the proposed Physics based informed SPF algorithm for RUL prediction of LiBs framework is presented by simultaneously considering multiple degradation mechanisms. This includes losses of active materials of the positive and negative electrodes and the loss of lithium inventory. The proposed approach uses a half-cell model to estimate degradation parameters from voltage and capacity measurements. This allows for quantifying the degradation mechanisms and predicting the capacity fade trend based on the estimated parameters. Unlike traditional capacity-based prognostics, which rely solely on the empirical capacity fade trend, the proposed approach offers a more comprehensive and accurate way to predict battery RUL. To ensure a reliable framework, accurately and rapidly estimating the initial parameters of the degradation model is crucial to prevent the gradient error from increasing during the prediction process. With this consideration, this thesis proposes a new hybrid approach that integrates data-driven and model-based approaches to enhance the accuracy of online prognostic health management prediction within the current framework. The proposed framework employs a Neural Network (NN) to model and monitor battery degradation trends while also determining the initial values of the degradation model under varying operating conditions. The results show that the proposed hybrid framework is more accurate and improves the convergence rate compared to the traditional capacity prognostic framework.
| Date of Award | May 2023 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Maher Al-Greer (Supervisor), Michael Short (Supervisor) & Victor Chang (Supervisor) |