Data-driven prediction of battery cycle life before capacity degradation

Introduction Precise estimation of battery cycle life is crucial for assessing quality and planning long-term battery management systems. Given the nonlinear nature of battery degradation, early and accurate cycle life prediction amid minimal degradation is exceptionally difficult. Machine learning (ML) methods, independent of specific mechanisms, can offer promising alternatives for predicting battery lifespan1. Li metal anode battery (LMB) is a promising next-generation battery because of the high energy density, whereas the safety concern with complex degradation mechanisms is still a crucial issue. For such challenge, ML-based cycle life prediction may play a role, though the LMB data are still limited. Here, we report our study on lifespan prediction of lithium metal-based batteries with high mass loading Ni-rich NMC electrode (NMC811/Li) by ML method through more battery features from various processes in the experimental charge/discharge cycles. We also explore the most crucial feature of NMC811/Li metal anode batteries' cycle life prediction with the feature importance analysis. Methodology In our project, 48 batteries are used to build ML model, 35 features are generated from the processes during cycle test, which are divided into discharge, charge, and voltage relaxation related features. The whole dataset is regarded as the training dataset, the testing datasets are generated by 4-fold cross validation of the whole dataset. ElasticNet and XGBoost ML methods are used in our study to predict the cycle life. Pearson’s correlation coefficient is used to define the relationship between features and cycle life. Feature importance analysis is done by XGBoost ML model under exhaustive feature selection method. Results and Discussion Firstly, we use discharge related features, charge capacity related features, and voltage relaxation features to predict the battery cycle life separately by ElasticNet linear regression model. We then find that discharge-related features give us the best prediction result compared to the other two feature sets. However, the prediction result is still not satisfactory. Hence, by calculating Pearson’s correlation coefficient we selected 12 features out of the 35 features which have strong or moderate correlation to cycle life. Through exhaustive feature selection, we generated totally 4095 feature subsets from the 12 features. We implement advanced XGBoost ML model on all the possible feature subsets to predict the cycle life. The result shows that XGBoost ML model has a significant increase of the prediction performance and there is one case which contains 6 features out of the 12 features can give us the best prediction result (Fig. (a)) which decrease the RMSE to 8.29, MAE to 6.45 and increase R2 to 0.89. In addition, we find logarithm of the minimum value of discharge capacity difference between 100 cycle and 10 cycle (Log(|min(100-10(V))|)) is the most important feature ranking for the 6 features in lifetime prediction (Fig. (b)). It contributes more than 44% for lifetime prediction. Finallly, we provide our model with 8 new NMC811/Li metal batteries as unseen data to test the prediction performance, The prediction result is shown in Fig. (c), the MAE and RMSE of the testing data are relatively small and the test error (Mean Absolute Percentage Error) equals to 6.6%, which suggests the model provides predictions with reasonable accuracy. This study underscores the successful application of ML in predicting cycle life for LMBs with high energy density design, paving the way for enhanced battery performance and longevity in practical applications. References Severson, Kristen A., et al. Nature Energy4(5) (2019): 383-391. Figure 1

Two-stage aging trajectory prediction of LFP lithium-ion battery based on transfer learning with the cycle life prediction

This paper mainly research the relationship between the lithium-ion battery management systems and charging policies and temperature.In the introduction of lithium-ion battery characteristics of the premise, the lithium-ion battery protection methods are described, based on a lithium-ion battery protection method in detail its charging policies and to study the effect of temperature on charging policies, compared with good interpretation of the lithium-ion rechargeable battery management system strategy selection and temperature control importance of the lithium-ion battery protection, providing a theoretical basis and ideas for research and specific battery management system.

Predicting the impact of formation protocols on battery lifetime immediately after manufacturing

Cycle life prediction of lithium-ion batteries based on data-driven methods

Accurate prediction of lithium-ion (Li-ion) battery cycle life using early cycle data is a challenging task as the capacity fade resulting from the nonlinear degradation process leads to a negligible loss of capacity in early cycles but is accelerated when approaching the end of life. To address this challenge, we propose a hybrid machine learning model that combines a shallow learning model, relevance vector machine (RVM), and a deep learning model, convolutional neural network (CNN). RVM is employed to generate artificial cells with high cycle lives, expanding the original training dataset (i.e., data augmentation). The expanded training dataset then serves as the input-output pairs used to train the CNN model. CNN first learns the locally-invariant features from the input data and then makes full use of these features to predict the cycle life of a Li-ion battery cell. We evaluate the performance of the proposed hybrid machine learning (RVM+CNN) model on two test datasets consisting of 83 cells with widely varying cycle lives ranging from 150 to 2300 cycles. The RVM+CNN model produces higher cycle life prediction accuracy on both datasets than three other machine learning and deep learning methods.

The lithium-ion battery has been widely used as an energy source. Charge rate, discharge rate, and operating temperature are very important factors for the capacity degradations of power batteries and battery packs. Firstly, in this paper we make use of an accelerated life test and a statistical analysis method to establish the capacity accelerated degradation model under three constant stress parameters according to the degradation data, which are charge rate, discharge rate, and operating temperature, and then we propose a capacity degradation model according to the current residual capacity of a Li-ion cell under dynamic stress parameters. Secondly, we analyze the charge and discharge process of a series power battery pack and interpret the correlation between the capacity degradations of the battery pack and its charge/discharge rate. According to this cycling condition, we establish a capacity degradation model of a series power battery pack under inconsistent capacity of cells, and analyze the degradation mechanism with capacity variance and operating temperature difference. The comparative analysis of test results shows that the inconsistent operating temperatures of cells in the series power battery pack are the main cause of its degradation; when the difference between inconsistent temperatures is narrowed by 5 °C, the cycle life can be improved by more than 50%. Therefore, it effectively improves the cycle life of the series battery pack to reasonably assemble the batteries according to their capacities and to narrow the differences in operating temperature among cells.

To advance the lithium-ion battery (LIB) technology more quickly, its lifetime should be predicted accurately. The precise prediction of LIB lifetime can help in producing new batteries, better use and operation of batteries. It is worthy for noting here that the LIB is a heavy nonlinear system suffering from battery fading, degradation, uncertainty and variability of operating conditions. Therefore, this article presents a hybrid extended Kalman filter with Newton Raphson method for lifetime prediction of lithium-ion batteries. The data analyses are based on commercial lithium iron phosphate/graphite cells cycled at fast charge. The cycle life expectancy is in the range of 150 to 2,300 cycles. The discharge voltage characteristics are used to present capacity degradation. The battery datasets are used with a hybrid Extended Kalman Filter (EKF) and Newton Raphson method to match the predicted cycle life and the actual cycle life of the battery. The effectiveness of the proposed method is verified by making a fair comparison with the linear regression-based machine-learning method. In the testing of 100 lifecycles, the test error and root mean square error record 3.26% and 10.93 compared with the linear regression that achieves 9.1% and 211, respectively. With the proposed hybrid approach, the lifetime prediction of LIBs can be further enhanced.

Rechargeable Lithium-ion batteries (LIBs) are the power source used widely in portable electronic devices and are becoming increasingly important in the automotive industry applications. LIBs need to be high-energy, safe, and cost-effective to address the demand for large driving ranges of batteries in electric vehicles (EVs) and hybrid electric vehicles (HEV). Therefore, technologies for extending the cycle life of LIBs and solving component degradation problems are required. Recently, the recharging time for batteries in EVs is longer compared to the refueling time for gasoline-powered vehicles which prevents the widespread adoption of battery-powered EVs into the transportation sector. In that perspective, achieving fast charge has been considered as one of the most important directions for the progression of EVs in the market and it can possibly lead to increased customer acceptability by reducing charging time. However, fast charging of LIBs makes them susceptible to lithium plating and induces high temperatures, and the decomposition of electrolytes results in thermal runaway and explosion. Thus, understanding capacity fades and improving cell stability for long cycle life at a high C-rate are critical for designing LIBs. It has been known that external pressure has a positive influence on Li-ion pouch cell capacity fading during aging experiments. Nonetheless, the effect of external pressure under high C-rate conditions is not well understood and has not been extensively covered in many studies. Even in some of the research conducted experiments under high C-rate conditions, the concept of normalized capacity has been introduced, leading to studies that do not account for the initial capacity drop [1,2].This work analyzed the effect of external pressure on the battery’s sustainability under high C-rate conditions. The interrelationship of mechanical pressure and the electrochemical performance was investigated using 1.05 Ah graphite/NMC622 stacked LIBs. The charging speed was set to 1,5 and 10C-rates in the constant current/constant voltage (CC-CV) mode to test fast charging conditions. The pressure and thickness changes were analyzed using the pressure measurement system and 3-dimensional digital image correlation (3D-DIC). Also, the post-mortem analysis conducted to confirm the morphology of each cell component and internal resistance under fast charging conditions were analyzed using scanning electron microscopy (SEM) analysis and electrochemical impedance spectroscopy (EIS), respectively. The results demonstrate the negative influence of the external pressure was confirmed initial cycling stage increases the internal resistance which lowers the ionic conductivity. Overall, we propose that innovative pressurizing strategies to maximize cell cyclability and pressure sensing results could provide a non-destructive diagnostic approach to support the advancement of fast-charging battery technologies. This research highlights the significance of external active pressurization in systems operating at high C-rates. Reference [1] Mussa, A.S., et al., Effects of external pressure on the performance and ageing of single-layer lithium-ion pouch cells. Journal of Power sources, 2018. 385: p. 18-26[2] Chen, F., et al., Air and PCM cooling for battery thermal management considering battery cycle life. Applied Thermal Engineering, 2020. 173: p. 115154 Figure 1

Lithium-cobalt oxide has become a new generation of highly promising anode materials for lithium-ion batteries due to its low price, environmental friendliness, high platform voltage, and high theoretical capacity. In this paper, the working characteristics and related parameters of lithium-ion batteries are sorted out, and the influence factors of the decline mechanism of lithium-ion batteries are investigated from the perspective of chemical composition. Regarding the cycle life of lithium-ion battery cathode materials, this paper establishes a cycle life prediction model for lithium-ion batteries based on the LSTM model. It optimizes the hyperparameters of the model using the PSO algorithm. In addition, this paper also prepared lithium cobalt oxide cathode materials and carried out a validation analysis of the factors affecting their electrochemical performance. It is found that the cycle life prediction of lithium-ion battery based on LSTM has an RMSE of 3.27%, and the capacity of lithium cobalt oxide soft pack full battery decays from 249.81mAh to 137.04mAh at 26°C. The cycle life of its lithium cobalt oxide lithium-ion battery is around 250 cycles, and the average decay is 0.445mAh during one charge/discharge. Different charge/discharge cycles and diversity will have a significant effect on the cycle life of lithium-ion battery cathode materials, and it is necessary to pay attention to the parameter changes in the preparation of lithium cobalt oxide cathode materials in order to increase the cycle life of lithium-ion batteries.

Problem. The article is devoted to the problem of increasing the safety, environmental friendliness and efficiency of vehicles through the use of lithium-ion batteries, predicting their final service life using a new predictive model of capacity degradation. The analysis of the performance and degradation of lithium-ion batteries is carried out and the factors of their degradation are studied. A predictive model of the degradation of the capacity of lithium-ion batteries in electric vehicles has been developed, which determines the remaining useful life of the battery and predicts its life cycle using data only from early charge/discharge cycles, during which significantly less degradation occurs. Methods for increasing the service life of electric vehicle batteries are given. Goal. The aim of the work is to improve the safety, environmental friendliness and efficiency of electric vehicles by determining the predicted final resource of lithium-ion batteries using a new predictive model of capacity degradation. Methodology. Methods of scientific analysis and synthesis of increasing the service life of batteries. A predictive model of capacity degradation of lithium-ion batteries of electric vehicles determines the remaining useful life of the battery and predicts its life cycle. The results. Based on the analysis of publications and studies, a predictive model of capacity degradation of lithium-ion batteries was developed. As a result, the final resource of lithium-ion batteries is predicted. An important aspect of the concept of predicting the state of degradation of the battery capacity is that the data analysis in combination with the results of this study demonstrates a virtually linear relationship between the life cycle, inflection point and curvature point. Methods for increasing the service life of batteries of electric vehicles and prolonging their life cycle through the rational use of an electric vehicle are considered. The results of the research coincide with the developed predictive model of lithium-ion battery capacity degradation, which determines that at each charge/discharge cycle, the electric vehicle battery loses an average of 0.015 % of its capacity over the entire life cycle. The considered methods for increasing the battery life, which allows increasing the number of charge/discharge cycles and the life cycle due to the rational use of the electric vehicle. Originality. The peculiarity of the developed predictive model of lithium-ion battery capacity degradation in electric vehicles is that the battery life cycle is determined by using data from the first charge/discharge cycles (from 200 to 250), where minor degradation still occurs. But such data will be sufficient to identify the distortion point, then the inflection point and then determine the full service life of the electric vehicle battery, which is limited to 80% of the useful capacity. Practical value. A predictive model for the capacity degradation of lithium-ion batteries in electric vehicles determines the remaining useful life of the battery and predicts its service life, which is an important issue for both electric vehicle owners (both new and used) and electric vehicle manufacturers (for the formation of warranty obligations and battery operation strategies).

Early prediction of lithium-ion battery cycle life based on voltage-capacity discharge curves

The application of lithium ion batteries (LIBs) have been widen from IT devices to electric vehicles (EVs). To be precise, EVs adopting LIBs are being increased because LIBs guarantee high energy density and cycle life. LIBs are also sensitive to external environmental factors such as temperature, humidity, vibration, etc. In order words, these conditions should be controlled for better battery performance and reliability. Most of the studies on LIBs have been focused on temperature as the key environmental factor, however, other variables such as humidity should be also considered carefully for EV application. Hence, this work focuses on the effect of humidity on self-discharging and battery degradation behaviour. Herein, we try to unveil the effect of relative humidity on self-discharging behaviour of fully charged LIBs by monitoring open-circuit voltages (OCVs), capacity and power retention, and electrochemical impedance before and after storage. Pouch-type LIBs having LiNi1/3Co1/3Mn1/3O2/graphite chemistry are fabricated, and then stored at room temperature under high humidity condition (>90%). As a result, the OCV drop is much higher at high humidity condition, which causes more degradation of capacity and power than expected. This abnormal result will be discussed in more detail via various case studies and analysis.

Surface fluorinated graphite suppressing the lithium dendrite formation for fast chargeable lithium ion batteries

Molten salt electrochemical upcycling of CO2 to graphite for high performance battery anodes