Battery Lifetime Analysis for Energy Harvesting Networks Steven Butt University of California, Los Angeles Electrical Engineering Department [email protected] Shane Erickson University of California, Los Angeles Electrical Engineering Department [email protected] ABSTRACT The conception of wireless distributed systems has attracted enormous attention. One of the largest hurdles standing in the way of their widespread deployment is their inability to store sufficient energy for long-term usage. Energy harvesting is a step in the right direction to solving this issue. Understanding the relationship between the energy harvested in the environment by a node and the actual energy stored in its battery is essential to ensuring reliable and energy efficient networks. To address this, a method of charging different batteries at different currents was employed. The corresponding energy stored for each charging current is measured by discharging each battery uniformly. The results demonstrate that there is little exploitable information between the way you charge a battery and how much energy the battery stores. Categories and Subject Descriptors B.8.2 [Performance Analysis and Design Aids]: Battery performance characteristics. General Terms Management, Measurement, Performance, Experimentation. Keywords Battery, Lifetime, Charging, Energy Harvesting. 1. INTRODUCTION Independent distributed systems, such as wireless sensor networks, are thought to someday play an integral part of our everyday lives. They might help scientists collect data in areas not easily accessed, adjust light or heat levels in a building to maximize energy efficiency, or warn us in an emergency. These systems, however, could have high maintenance costs if batteries are used to power them since they will have to be replaced periodically. The larger the network, the greater the resource costs required to maintain it. To combat this issue, the concept of energy harvesting has arisen. The idea suggests that nodes in a network, whether stationary or mobile, should have the ability to convert some form of environmental energy (e.g. solar, magnetic, heat, etc.) into useable or storable energy that can be used by that node. In the case of mobile nodes, they might be capable of delivering energy to nodes that do not have the capability to harvest it themselves. Regardless, if a network has the ability to gather energy from the environment and store more than it can consume, the network may vary rarely need outside maintenance. This paper will address the issue of storing energy via batteries, specifically 1.2V rechargeable NiMH batteries since this category of battery is currently used in wireless sensor network devices. It will also seek to discover the relationship between the rate at which you charge a battery and the effective energy stored. The difficulty here lies in the non-idealities of batteries. Implications of how this information can be used in wireless sensor networks will also be addressed. 2. PREVIOUS WORK Ideally, we would like a battery to supply a constant voltage during its lifetime. Similarly, we would also expect that when recharging a battery, the amount of current we supply is equivalent to the amount we can extract at a later time. These idealities are unfortunately unrealistic. The voltage of a battery changes during discharge, and further variances occur based on the discharge rate and temperature [2]; larger discharge rates lower the lifetime while higher temperatures increase it. A number of quantitative models exist that predict the lifetime of a battery based on characteristics of the battery and its load profile [5]. Memory Effect is another interesting property of batteries. This concept describes how charging and discharging batteries can affect lifetime based on the level you charge and discharge it to. As described in [2], a fully charged and operational 1.2V rated NiMH battery is discharged to 1.15V and then fully charged again. This process is repeated eighteen times and the lifetime of the battery has decreased almost 20%. The battery is then fully discharged to 1.0V and fully charged. After three iterations of this, the battery has achieved a lifetime almost equal to that of the initial fully charged battery. Despite the well-understood ideas behind discharging batteries, how discharging at different rates and temperatures affects battery lifetime, and how memory effect fits into the equation, there is still little to no understanding on how to predict the lifetime of a battery based on how long it has been charged and at what rate. Ideally then, we would like to develop a model for charging batteries similar to those established for discharging. We hypothesized that charging batteries at lower current rates would increase the useful lifetime of the battery. 3. EXPERIMENTATION 3.1 Software approach In order to test our hypothesis, we required an experimental setup that allowed us to test how much useful energy we could extractfrom a battery after charging it to some set amount of energy capacity. Initially we desired a software approach instead of a hardware approach. This would involve the use of a battery simulator called DUALFOIL created at UC Berkeley that accurately simulates battery chemistry over the course of a discharge cycle. Using this program would have provided a convenient and time-saving method to test the hypothesis. However, the problem with this program, and consequently the reason why we chose not to use it, is the fact that it does not simulate the process of charging a battery. This was the crux of our experiment and hence precluded us from using DUALFOIL to obtain experimental data. 3.2 Hardware Approach Since there was no available software to run our experimental platform on, we resorted to a hardware approach. Hardware experimentation has a few problems, and the first of which is the actual length of experimentation time. The process of charging a battery and then discharging it is a tedious process that takes many hours or even days for one result to be established. In the worst case we would have to charge a battery over the course of 50 hours, then take it out of the charging mechanism and discharge the battery for another 3-4 hours. In the best case, the process would take a quarter of a day. The time factor limited the number of data points we were able to obtain for the results section. A second implication