Online capacity estimation of lithium-ion batteries with deep long short-term memory networks

The challenge and opportunity of battery lifetime prediction from field data

Теорія ймовірності і математична статистика (укр.)Probability theory and mathematical statistics (en.)

With the increasing use of rechargeable lithium-ion battery packs in numerous applications, it calls for an effective evaluation of active battery cell equalization to enhance the whole battery pack's capacity and performance. Plenty of work has focused on cell equalizing circuit and control algorithm design. Still, none of them is devoted to a comprehensive analysis and comparison of cell balancing methods from the topology level. In this article, a review of the state-of-the-art active battery cell equalization methods is conducted, where it is classified as adjacent-based, nonadjacent-based, direct cell-cell, and mixed topologies. This classification can provide a comprehensive way to analyze and compare the existing active cell balancing methods’ performance. Subsequently, the mathematical models of these balancing structures are developed, and extensive simulation results are provided to compare the performance of these equalization topologies in terms of equalization speed. It then comes up with an inclusive economic comparison between them, followed by a discussion.

Active Cell Equalization Topologies Analysis for Battery Packs: A Systematic Review

To accurately estimate the battery state of health (SOH) is crucial to enhance the performance of a battery-powered system. This paper proposes a SOH estimation method that can predict the battery current SOH when the battery is fully charged. When the state of charge (SOC) is 100%, the AC impedance of the Li-ion battery at each frequency can be obtained by electrochemical impedance spectroscopy (EIS) and used as the training data of the neural network. The result has been validated by another aging battery and it shows that with the best performance neural network, the maximum relative error of the estimated SOH is only 1.31 %. Besides, compared to the linear interpolation method, the maximum relative error has improved by 1.74 %, MRE by 0.50 %, MAE by 0.42 %, and MSE by 0.86 %. Furthermore, this paper discusses a technique that can estimate the battery SOH under different temperatures without collecting the AC impedance through the whole cycle life test. • An active SOH estimation method has been established by ANN. • AC impedance of the battery is utilized as the training data of ANN. • Removing the non-linear data reduces the scale of ANN and improves its performance. • The MAE, MRE and MSE of proposed ANN are 0.35, 0.41 % and 0.21 %, respectively. • The sensitivity of the neural network to the temperature has been validated. 

An online state of health estimation technique for lithium-ion battery using artificial neural network and linear interpolation

Usable capacity refers to the maximum capacity in theory that a fully charged battery can release, and is often used as an indicator in state of health (SOH) estimation for lithium ion batteries. The traditional method for measuring usable capacity is mainly based on voltage data in the discharge process with a constant current. However, the discharge current of a lithium ion battery in operation always fluctuates due to load changes, which makes the traditional method difficult for realizing online capacity measurement. To overcome the above problems, a novel approach is proposed in this paper to estimate the usable capacity and SOH of lithium ion batteries based on the charge curve. The time intervals between two voltages and currents during charging are used as the health factors to predict the usable capacity, which is then used to perform the SOH estimation. Experiments are implemented based on data provided by the NASA Ames Prognostics Center of Excellence. Results confirm that the proposed method performs well in online estimation of SOH.

Online estimation of state-of-health for lithium ion batteries based on charge curves

Field data is often used as the basis for the prediction of an airplanes maintenance request In the traditional methods, maintenance request predictions are mainly obtained immediately using data However, uncertainty analysis during the failure detection is ignored, which makes maintenance request inaccurate To overcome the above problems, a novel approach is proposed in this paper: a multivariate damage model is established to obtain the degree of airplane damage, which is used as an indicator for maintenance request predictions On the basis of the degree of damage, uncertainty analysis can be effectively described using a stochastic process and the Markov process The transition probability and transition time corresponding to the potential detection rate and date of maintenance, which are used to determine the distribution of maintenance requests Experiments are implemented based on field data of a certain type of airplane Results confirm that the proposed method performs well in the predictions of maintenance requests

Maintenance request prediction for airplanes based on multivariate damage model

In the field of defense science and technology, some long-life and high-reliability products cost a lot to carry out reliability test, including life test, degradation test and enhancement test. Thus, a more effective product testing and data analysis method are expected. Reliability enhancement test (RET) is commonly employed to improve products reliability, whose data is usually utilized to determine the working stress limit and damage limit. However, data from RET may contain much more information about products such as degradation process and stress influence information. The stress effect can be divided into: stress acceleration effect and stress compensation effect. The existing research often ignores this information. A modelling method scheme based on different enhancement data characteristics is proposed in this paper. Furthermore, an accelerated-compensated degradation modeling method for the data with degradation-acceleration-compensation data is developed to further utilize enhancement test data and lay the foundation for the subsequent reliability evaluation. At last, the RET data set from the servo system of a typical equipment is adopted to validate the method proposed previously.

Nina Yang

Papers

Online estimation of state-of-health for lithium ion batteries based on charge curves

Maintenance request prediction for airplanes based on multivariate damage model

Accelerated-compensated degradation modeling based on synthesized-stress enhancement test data