Introduction

Nowadays it is almost impossible to imagine a modern person without a smartphone. A smartphone is needed to communicate with family and friends, for entertainment, education and much more. The smartphone market is quite competitive and at the moment has more than a hundred brands, the number of which is only growing every year. Each brand tries to stand out among the others with its own characteristics and despite the variety of components in smartphones, most manufacturers use the same technology for batteries – lithium-ion. Many users are not even aware of the fact that it is the battery in a smartphone that wears out the fastest. Therefore, the task of predicting the lifetime of a lithium-ion battery is critical, given its widespread use, but difficult because of nonlinear degradation and wide variability, even when operating conditions are met [1-5].

Approaches using statistical methods and machine learning techniques to predict lifetime are attractive alternatives, independent of mechanism. Recent advances in computational power and data generation have accelerated progress on a variety of tasks, including predicting material properties [6-7], determining chemical synthesis routes [8], and finding materials for energy storage [9-11]. Accurate early lifetime prediction with much less degradation is a challenge due to the typically nonlinear degradation process (with little loss of capacity in early cycles), and the relatively small datasets used today, which cover a limited range of lifetimes [12]. Machine learning methods are particularly attractive for high-speed operating environments, where primary theoretical degradation models are often unavailable. In other words, opportunities for improving current prediction models include higher accuracy, earlier prediction, and greater interpretability.

Goals and tasks of the research

The goal of the research is to develop an approach to predicting the remaining lifetime (number of charge/discharge cycles until complete wear and tear) of a battery under different charging conditions based on machine learning, which should take into account the advantages and disadvantages of already existing prediction methods.

The tasks of the research consist in the analysis of processes occurring in power supplies of mobile devices; in the analysis of advantages and disadvantages of existing methods to improve the efficiency of power supplies for mobile devices; in the study of analytical and simulation models of power supplies for mobile devices; in finding possible ways to improve existing models of power supplies for mobile devices; in developing its own improved model of processes occurring in power supplies for mobile devices, as well as evaluating the effectiveness of the proposed model.

Review of research and development

Most of the discussions and developments on this topic are conducted in the English-speaking space. In articles on a similar topic, more and more attention is now paid to improving the efficiency of electric car batteries, as this industry is developing quite rapidly as many developed countries are trying to reduce the level of hydrocarbon emissions into the atmosphere.

Review of international sources

In the article A comprehensive investigation of lithium-ion battery degradation performance at different discharge rates [13] studied the behavior of a lithium-ion battery at different discharge rates. The battery behavior identified in this article can help manufacturers and consumers better understand battery properties and provide recommendations for battery optimization and application. As a key methodology, the dynamic Peukert's [14] law proposed by the authors of this article can be extended to many other research problems, such as battery degradation modeling, condition monitoring, and problems in predicting the remaining lifetime under varying workload conditions. Another problem worth considering in the future is the mechanism underlying Peukert's law and the corresponding electrochemical degradation behavior explaining the dynamic Peukert's law.

In the article Numerical simulation of prismatic lithium-ion battery life cycles under a wide range of temperatures [15] the authors developed an electrochemical and thermochemical numerical model. Using the numerical model, the lithium-ion battery was investigated at various operating temperatures. As a result of the study, the performance and lifetime of the lithium-ion battery was successfully described. It follows from the results that the performance degradation of the lithium-ion battery is 40% at an operating temperature of 0°C. Numerical simulations of energy storage capacity during cycling predict that increasing the temperature above 20°C significantly reduces capacity retention, which is critical to the long life of the lithium-ion battery. With proper thermal management, the temperature of the lithium-ion battery should be controlled at 20°C for maximum performance and longevity. Therefore, a thermal management strategy in the lithium-ion battery is essential for maximum battery utilization, longevity and safety.

In the article Lithium-ion battery life prediction based on initial stage-cycles using machine learning [16] proposed a lithium-ion battery life prediction model using discharge data at the beginning of the cycle. The proposed model allows increasing the learning rate while obtaining results comparable to traditional approaches. The work done by the authors of this paper underscores the promise of combining data generation and modeling based on this data to understand and design complex systems such as the lithium-ion battery. The proposed approach for predicting the remaining lifetime of a lithium-ion battery in the context of different battery operating conditions can be improved. In the future, the proposed model can be used in applications that can accelerate research and development of new battery designs, as well as reduce production time and cost. In addition, the model can reduce the time to test new battery types, which is especially important given the rapid advances in materials.

Review of national sources

In the article Modeling of batteries and their assemblies taking into account the degradation of parameters [17] the main methods and directions of modeling the life cycle of the battery are considered. The developed model was used to calculate a parallel-sequential assembly of four batteries, and the deficit of the electric capacity of one battery in the assembly on the operation of the assembly as a whole was also investigated. Reducing the capacity of the defective battery by 10% resulted in a 2.9% reduction in the discharge time of the studied assembly compared to the discharge time of the reference assembly.

In the article Simulation and estimation of lithium-sulfur battery charge state using fuzzy neural network [18], a neural fuzzy adaptive network (ANFIS), which is a highly specialized self-learning expert system, is proposed to assess the charge state during the discharge of a lithium-sulfur battery. The developed ANFIS model estimates the charge state of a lithium-sulfur battery during its galvanostatic cycling with sufficient accuracy (the error is less than 5%). However, the developed model allows estimating the state of charge only at constant values of charging and discharging currents. Further development of a more complex model that takes into account the influence of cycling modes (charging and discharging currents, pauses, temperature and other parameters) on the charging state of the lithium-sulfur battery is required. To develop such a model, experimental data on cycling of the lithium-sulfur battery in different modes are needed.

Review of local sources

Among the works of DonNTU masters, published on their personal websites, we can highlight the work of Mirnyi A. (2018 graduate), who in his article Analysis of the combined autonomous unit system considered the advantages of lithium-ion batteries and argued the reasons for their use in fixed combined autonomous units, and the work of Temertey N. (2018 graduate), who in her master's thesis considered the chemical processes that take place in lithium-ion batteries; the principle of operation of lithium-ion batteries; the advantages and disadvantages of lithium-ion batteries.

General information about lithium-ion batteries

Lithium-ion batteries are used not only in smartphones, but also in electric cars, TVs, drones, and other devices. Consequently, the question of making the most of this technology is more relevant than ever. To begin with, it is worth noting that there are different types of batteries for mobile devices, the main difference between which is the type of active substance. The most common types of active substance for mobile devices are lithium-ion (Li-Ion) and lithium-polymer (Li-Po, Li-Pol, Li-Poly).

An important feature of lithium-ion batteries is the short charging time, which in some cases can be up to 20 minutes [19]. Some manufacturers of these batteries advise charging them at about 80% to extend their lifetime. The charging efficiency under such conditions will be about 99%, and the temperature change during charging will be negligible and will not have any negative consequences. Some types of lithium-ion batteries can tolerate temperature rises of up to 5°C when fully charged. This is usually due to increased internal resistance or to protective circuitry. When the battery reaches a threshold voltage, it is fully charged. Fully charging the battery is not recommended because the high voltage causes an unbalanced battery. Lithium-ion batteries also tend to self-discharge. This means that if you neglect proper operation, the battery will lose about 0.5-1% of its maximum capacity per month.

Identification of characteristics influencing the cycle life of lithium-ion batteries

To begin with, a few key points should be highlighted. There is a state of charge (SOC), which characterizes the degree of battery charge (100% – fully charged, 0% – fully discharged) and the equivalent depth of discharge (DOD), which is calculated by the equation:

DOD = 100% - SOC

(1)

The self-discharge rate is usually measured by measuring the decrease in open-circuit voltage (OCV) over time. The OCV is the voltage between two pins of an electrical circuit when the load connected to those pins is disconnected. The difficulty of such measurements is not in the complexity, but in the amount of time needed to detect this process. However, studies have been carried out on this subject [20], which revealed typical no-load voltage dependencies on the state of charge. In fact, the slow dynamics comes down to a simulation of the effect of the SOC on the electrical characteristics of the battery. The no-load voltage (OCV) was found to be a fairly unambiguous function of the state of charge (SOC or DOD). This dependence is presented in Figure 1.

A study by Candler Smith, Aron Saxon, Matthew Keiser, and Blake Lundstrom [21] suggests that the life of lithium-ion batteries will vary depending on their thermal environment and how they are charged and discharged. In their study they used eleven 75 ampere-hour (Ah) batteries from Kokam, for comparison, modern smartphones have a capacity of about 4,000 milliampere-hours (mAh), which equals 4 Ah. These eleven batteries were tested under nine different wear conditions, as shown in Table 1.

Table 1 – Tests wear for batteries company Kokam

The cells were fully charged at a constant current of up to 4.2 V (Volts), until the current decreased to less than C/10. C is the current depending on the capacity of the battery (1C = 100% of the capacity), e.g. if the battery capacity is 4000 mAh, then 1 C = 4 A. The cells were fully discharged at constant current to a minimum voltage of 3.0 V. The maximum voltage range of 4.2 V/3.0 V for the 100% DOD wear tests was narrowed to 4.1 V/3.4 V for the 80% DOD wear tests.

All wear tests were interrupted once a month for a reference performance test. All reference tests were conducted at specific temperatures, except for Cell 11, which was lowered to 45°C to comply with the manufacturer's temperature limits during charging. At room temperature, cells 1 and 2 showed the minimum wear cycle. Cells 6 and 7, held at 0°C and experiencing severe fading, showed about a 10 percent difference in fading rate. Because of this information, we can assume that temperature has a fairly strong effect on the rate of fading. Inside batteries is the electrolyte, the temperature of which affects the mobility of ions, the speed of chemical reactions, and overall performance. Different types of power sources have different electrolyte compositions and therefore react differently to negative temperatures, therefore, the above study will only be valid for lithium-ion batteries.

Conclusions

Data-driven modeling is a promising way to predict lithium-ion battery life. Approaches in developing battery life prediction models can be complemented by approaches based on physical and semiempirical models as well as specialized diagnostics. Overall, research in this area underscores the promise of combining data generation and modeling based on that data to understand and develop complex systems such as lithium-ion batteries.

References