< Terug naar vorige pagina
Publicatie
Developing prognostics and health management methodology for lithium-ion battery cells
Boek - Dissertatie
Lithium ion batteries have been broadly used as energy storage systems (ESS) in electric powertrains such as electric vehicles. The assessment of battery life is a critical element in the success of ESS since battery failure imposes performance inefficiency, augmented degradation, and expensive maintenance. Therefore, due to strict needs for the reliability of ESS and to satisfy the cost constraints, a reliable maintenance strategy should be pursued. The adopted strategy should provide not only accurate, early, and online health diagnosis but also predict the remaining useful life (RUL) of the lithium battery, with high precision, within a confidence interval, and independent of the operating conditions. Battery prognostics and health management (PHM) is an advanced solution that has emerged to optimize battery maintenance and to ensure that the battery can satisfy the power and energy demand of its application. This Ph.D. dissertation focuses on developing novel and promising methods for an online PHM of lithium ion batteries using intelligent techniques such as artificial intelligence (AI) methods.
First, a comprehensive review of prognostics health management (PHM) methodology for Lithium ion batteries using machine learning techniques is conducted. This literature survey provides a comprehensive insight into PHM modules from data acquisition to health diagnostics and prognostics models. It presents the most common data acquisition and data processing techniques that are applied for the health diagnosis and prognosis of Lithium ion batteries. Moreover, it focuses on studying various types of battery health indicators and classifying them based on their characteristics. Lastly, different health diagnostics and prognostics methods are introduced, and their characteristics are compared. This comprehensive survey gives insight into existing methods, the limitation of previous researchers, and the gaps in the fields of battery PHM.
In this dissertation, an online PHM model for Lithium ion batteries is developed. It includes four main steps of data acquisition, feature engineering, health diagnosis, and health prognosis.
In the data acquisition step, a comprehensive battery experiment was conducted on two types of nickel manganese cobalt (NMC) cells. The next step is health diagnosis; Herein, the partial charging voltage curve is explored to extract the health indicators that describe the health trajectory of the battery. Afterward, a type of recurrent neural network method called nonlinear autoregressive model with exogenous input (NARX) is trained to capture the dependency between the health indicators and state of health of battery cells. The output of this step is battery SOH which will be used in the health prognostics step. In the health prognostics step, a predictive model called the pairwise similarity-based approach is developed to predict the RUL of the battery. Illustrative results demonstrate that the proposed PHM technique can estimate the SOH and predict the RUL of tested battery cells with high precision and low computation cost.
Finally, motivated by the fact battery health condition extremely depends on the application to which it has been subjected, this dissertation further exploits a model for health monitoring of lithiumion batteries under a real operating condition, in which a variety of stress factors can come into play. To this end, the battery cells have been cycled under a worldwide light-duty driving test cycle (WLTC) load profile to acquire real-world driving data. The time and frequency domain health indicators are extracted from the voltage signal. The extracted health indicators are then fed into a Gaussian process estimator (GPR) to track the real-time state of health. Results reveal the proposed approach is highly precise and is capable of estimating battery SOH with low computational costs and a relative error of less than 1 %.
First, a comprehensive review of prognostics health management (PHM) methodology for Lithium ion batteries using machine learning techniques is conducted. This literature survey provides a comprehensive insight into PHM modules from data acquisition to health diagnostics and prognostics models. It presents the most common data acquisition and data processing techniques that are applied for the health diagnosis and prognosis of Lithium ion batteries. Moreover, it focuses on studying various types of battery health indicators and classifying them based on their characteristics. Lastly, different health diagnostics and prognostics methods are introduced, and their characteristics are compared. This comprehensive survey gives insight into existing methods, the limitation of previous researchers, and the gaps in the fields of battery PHM.
In this dissertation, an online PHM model for Lithium ion batteries is developed. It includes four main steps of data acquisition, feature engineering, health diagnosis, and health prognosis.
In the data acquisition step, a comprehensive battery experiment was conducted on two types of nickel manganese cobalt (NMC) cells. The next step is health diagnosis; Herein, the partial charging voltage curve is explored to extract the health indicators that describe the health trajectory of the battery. Afterward, a type of recurrent neural network method called nonlinear autoregressive model with exogenous input (NARX) is trained to capture the dependency between the health indicators and state of health of battery cells. The output of this step is battery SOH which will be used in the health prognostics step. In the health prognostics step, a predictive model called the pairwise similarity-based approach is developed to predict the RUL of the battery. Illustrative results demonstrate that the proposed PHM technique can estimate the SOH and predict the RUL of tested battery cells with high precision and low computation cost.
Finally, motivated by the fact battery health condition extremely depends on the application to which it has been subjected, this dissertation further exploits a model for health monitoring of lithiumion batteries under a real operating condition, in which a variety of stress factors can come into play. To this end, the battery cells have been cycled under a worldwide light-duty driving test cycle (WLTC) load profile to acquire real-world driving data. The time and frequency domain health indicators are extracted from the voltage signal. The extracted health indicators are then fed into a Gaussian process estimator (GPR) to track the real-time state of health. Results reveal the proposed approach is highly precise and is capable of estimating battery SOH with low computational costs and a relative error of less than 1 %.
Aantal pagina's: 172
Jaar van publicatie:2022
Toegankelijkheid:Open