Search

The inevitable battery ageing is a bottleneck that hinders the advancement of battery-based energy storage systems. Developing a feasible health assessment strategy for battery pack is important but challenging due to the joint requirements of the computational burden, modeling cost, estimation accuracy, and battery equalization. This article proposes a balancing current ratio (BCR) based solution to achieve reliable state-of-health (SoH) estimations of all series-connected cells within a pack while significantly reduce the overall reliance on cell-level battery models. Specifically, after employing BCR to describe the properties of the balancing process, the voltage-based active balancing is combined into the SoH estimator design for the first time, leading to a weighted fusion strategy to effectively estimate SoHs of all cells within a pack. Hardware-in-the-loop experiments show that even if a parameter-fixed open-circuit-voltage-resistance model is used for modeling, the typical estimation error of our proposed solution can still be bounded by only 1.5%, which is 70% lower than that of the benchmarking algorithms. Due to the model-free nature of the integrated voltage-based balancing, the robustness and flexibility of the proposed pack SoH estimation solution are also significantly improved. [ABSTRACT FROM AUTHOR]

With the advent of sustainable and clean energy, lithium-ion batteries have been widely utilised in cleaner productions such as energy storage systems and electrical vehicles, but the management of their electrode production chain has a direct and crucial impact on the battery performance and production efficiency. To achieve a cleaner production chain of battery electrode involving strongly-coupled intermediate parameters and control parameters, a reliable approach to quantify the feature importance and select the key feature variables for predicting battery intermediate products is urgently required. In this paper, a Gaussian process regression-based machine learning framework, which incorporates powerful automatic relevance determination kernels, is proposed for directly quantifying the importance of four intermediate production feature variables and analysing their influences on the prediction of battery electrode mass load. Specifically, these features include three intermediate parameters from the mixing step and a control parameter from the coating step. After deriving four different automatic relevance determination kernels, the importance of these four feature variables based on a regression modelling is comprehensively analysed. Comparative results demonstrate that the proposed automatic relevance determination kernel-based Gaussian process regression models could not only quantify the importance weights for reliable feature selections but also help to achieve satisfactory electrode mass load prediction. Due to the data-driven nature, the proposed framework can be conveniently extended to improve the analysis and control of battery electrode production, further benefitting the manufactured battery yield, efficiencies and performance to achieve cleaner battery production. Image 1 • A data-driven framework is designed for mass load prediction in battery production. • Importance of four production feature variables of interest is directly quantified. • Performance of four ARD kernels is comparatively studied via GPR model. • Effects of these sensitive features on the mass load predictions are analysed. • Designed framework can benefit the optimisation for cleaner battery production. [ABSTRACT FROM AUTHOR]

Prediction of battery calendar ageing is a key but challenging issue in the development of durable electric vehicles. This paper simultaneously evaluates three mainstream types of modelling techniques for calendar ageing prediction of Lithium-ion (Li-ion) batteries. They are the pseudo two dimensional (P2D)-based electrochemical model, Arrhenius law-based semi-empirical model, and Gaussian process regression (GPR)-based data-driven model. Specifically, both the electrochemical and semi-empirical models are consciously developed or selected from the state-of-the-art modelling literature. For the data-driven model, due to the limited research in the existing publications, a machine learning-enabled GPR model is derived and applied for calendar ageing prediction. An experimental setup is developed to load the commercial Panasonic NCR18650BD batteries and to collect the experimental calendar ageing data under different storage temperature and SOC levels over 435 days. Based upon this well-rounded database, each model is well trained through using its corresponding training solution. Then the prediction performances of these models are studied and evaluated in terms of the model accuracy, generalization ability and uncertainty management. Both the challenges and future prospects of each model type are highlighted to assist the industrial and academic research communities, thus boosting the progress of designing advanced modelling techniques in battery calendar ageing prediction domain. • Accurate predictions of calendar ageing are crucial for battery real applications. • Three representative modelling techniques are evaluated based on a well-rounded database. • The accuracy, generalization ability and uncertainty management of these models are compared. • Benefits and key challenges of each model type are systematically analysed and discussed. • Inspiring future prospects are given toward the improvement of existing techniques. [ABSTRACT FROM AUTHOR]

Predicting batteries' future degradation is essential for developing durable electric vehicles. The technical challenges arise from the absence of full battery degradation model and the inevitable local aging fluctuations in the uncontrolled environments. This paper proposes a base model-oriented gradient-correction particle filter (GC-PF) to predict aging trajectories of Lithium-ion batteries. Specifically, under the framework of typical particle filter, a gradient corrector is employed for each particle, resulting in the evolution of particle could follow the direction of gradient descent. This gradient corrector is also regulated by a base model. In this way, global information suggested by the base model is fully utilized, and the algorithm's sensitivity could be reduced accordingly. Further, according to the prediction deviations of base model, weighting factors between the local observations and base model can be updated adaptively. Four different battery datasets are used to extensively verify the proposed algorithm. Quantitatively, the RMSEs of GC-PF can be limited to 1.75%, which is 44% smaller than that of the conventional particle filter. In addition, the consistency of predictions when using different size of training data is also improved by 32%. Due to the pure data-driven nature, the proposed algorithm can also be extendable to other battery types. • Design of a gradient-correction particle filter for battery aging prediction • Gradient corrector is employed to improve the tracking performance of particles. • Model-based regularization is proposed to reduce the algorithm sensitivity. • The consistency of predictions under different training data size is guaranteed. • Experimental results show that the prediction errors can be limited within 1.75%. [ABSTRACT FROM AUTHOR]

Accurate health estimation and lifetime prediction of lithium-ion batteries are crucial for durable electric vehicles. Early detection of inadequate performance facilitates timely maintenance of battery systems. This reduces operational costs and prevents accidents and malfunctions. Recent advancements in "Big Data" analytics and related statistical/computational tools raised interest in data-driven battery health estimation. Here, we will review these in view of their feasibility and cost-effectiveness in dealing with battery health in real-world applications. We categorise these methods according to their underlying models/algorithms and discuss their advantages and limitations. In the final section we focus on challenges of real-time battery health management and discuss potential next-generation techniques. We are confident that this review will inform commercial technology choices and academic research agendas alike, thus boosting progress in data-driven battery health estimation and prediction on all technology readiness levels. • Battery ageing mechanisms and the most common stress factors are discussed. • Data-driven technologies for battery SOH estimations are summarized regarding the benefits and drawbacks. • Data-driven health prediction methods including analytical models with data fitting, and machine learning methods are reviewed. • A compilation of the existing issues and challenges in this field is given. • Feasible and cost-effective solutions are suggested toward the improvement of related data-driven technologies. [ABSTRACT FROM AUTHOR]

Battery modeling and state estimation are critical to the battery safety and vehicle driving range. Currently, electrochemical mechanism based models require huge computation effort for solving partial differential equations, equivalent circuit models do not consider actual battery mechanism, while machine learning models are lack of generalization ability due to their data-driven nature. All these problems bring the challenges to guarantee the model effectiveness in wide temperature and large current conditions. To estimate the battery state of charge (SOC) and state of temperature (SOT) under these conditions, an electrochemical-thermal-neural-network (ETNN) model is formulated in this paper. Specifically, a simplified single particle model and a lumped thermal model are served as the sub-models of ETNN to predict core temperature and provide approximate terminal voltage. Then a neural network is incorporated to enhance the performance of sub-models. According to the extensive experiments, ETNN model is able to accurately estimate battery voltage and core temperature under the ambient temperatures of -10–40 °C and the discharge rate of 10-C. After that, an unscented Kalman filter (UKF) is integrated with ETNN to achieve reliable co-estimation of SOC-SOT. Experimental results illustrate that proposed ETNN-UKF can rapidly eliminate initial errors and provide satisfactory co-estimation performance. • Formulation of a hybrid battery electrochemical-thermal-neural-network model. • High model accuracy under large currents and wide range of ambient temperatures. • Accurate co-estimation of SOC and SOT by integrating the unscented Kalman filter. [ABSTRACT FROM AUTHOR]

Batteries are presently pervasive in portable electronics, electrified vehicles, and renewable energy storage. These indispensable engineering applications are all safety-critical and energy efficiency-demanding such that batteries must be meticulously monitored and manipulated, where effectively estimating the internal battery states is a key enabler. The primary goal of this paper is to present a concise, understandable overview of existing methods, key issues, technical challenges, and future trends of the battery state estimation domain. More specifically, for the first time, the state of the art in State of Charge (SOC), State of Energy (SOE), State of Health (SOH), State of Power (SOP), State of Temperature (SOT), and State of Safety (SOS) estimation is all elucidated in a tutorial yet systematical way, along with existing issues exposed. In addition, from six different viewpoints, some future important research opportunities and evolving trends of this prosperous field are disclosed, in order to stimulate more technologically innovative breakthroughs in SOC/SOE/SOH/SOP/SOT/SOS estimation. • Critical battery internal states are defined and explained. • Battery state estimation methods are systematically overviewed. • Key issues, technical challenges and opportunities are elucidated. • Future research directions and trends are presented. [ABSTRACT FROM AUTHOR]

This article presents a thorough comparative study of energy management strategies (EMSs) for a parallel hybrid electric vehicle (HEV), while the battery ageing is considered. The principle of dynamic programming (DP), Pontryagin's minimum principle (PMP), and equivalent consumption minimization strategy (ECMS) considering battery ageing is elaborated. The gearshift map is obtained from the optimization results in DP to prevent frequent shifts by taking into account drivability and fuel economy, which is then applied in the PMP and ECMS. Comparison of different EMSs is conducted by means of fuel economy, battery state-of-charge charge-sustainability, and computational efficiency. Moreover, battery ageing is included in the optimization solution by utilizing a control-oriented model, aiming to fulfill one of the main cost-related design concerns in the development of HEVs. Through a unified framework, the torque split and battery degradation are simultaneously optimized in this study. Simulations are carried out for DP, PMP, and ECMS to analyze their features, wherein results indicate that DP obtains the best fuel economy compared with other methods. Additionally, the difference between DP and PMP is about 2% in terms of fuel economy. The observations from analysis results provide a good insight into the merits and demerits of each approach. • Comparative studies of different energy management strategies are performed for a parallel HEV. • The fuel economy, drivability and battery ageing are concurrently considered. • The gearshift map is extracted considering drivability and fuel economy simultaneously. • A control-oriented battery ageing model is given to consider battery ageing dynamics into the optimization solution. • The performance for these methods under China's typical city driving conditions is analyzed. [ABSTRACT FROM AUTHOR]