Using digital technology, the functions of data
processing, performance logging and charging
control can optimize battery performance while
improving the accuracy of forecasted run times
and storage capacity.
For the PDF version of this article, click here.
Battery management has used digital control for about as long as cost-affordable microcontrollers have been available. There are 20-year-old patents from when microcontrollers were used to collect data and control charging. In 1994, the Smart Battery System (SBS) specification standardized the battery management interface and commands. This standardization helped foster microcontroller development with specific peripherals to measure battery characteristics and control battery activity.
As part of the SBS specification, the system and battery could communicate with a power supply used as a charger. This allowed the power supply to be configured for the correct charging voltages and to properly charge the battery under a set of given operating restrictions. These configuration parameters are relatively simple, and in many cases only define “limits.”
Battery management can certainly benefit from recent developments in digitally controlled power. Digitally controlled power can be used to charge the battery as well as convert power delivered by the battery. In addition, PMBus, the proposed industry-standard communication for digitally controlled power, shares a common transport layer, SMBus. Although the command language differs, this common transport allows for co-existence on the same physical bus.
Benefits of Digitally Controlled Power
Digitally controlled power extends power management from monitoring and configuring to high-resolution control of the switching duty cycle. These power-specific digital solutions have on-board CPU, memory and standard controller peripherals such as a general-purpose timers and ADC. In addition, there are specific peripherals that can be digitally controlled to address power applications such as very-high-resolution pulse-width modulators (PWMs). The information used to “close the loop” in a digitally controlled power system can also analyze the battery.
At the 2005 Digital Power Forum, one paper described a digital power system that uses a pseudo-random binary sequence (PRBS) for automated identification and control design. PRBS is used to slightly modulate the regulating duty cycle, and a controller then measures the output signal's perturbation to determine the power system's dynamics including load characteristics. This information is then used to optimize the power supply performance by modifying the control loop.
An example of this system is shown in Fig. 1. However, in this example, the battery's characteristics are analyzed to determine its impedance over a range of frequencies. The basic principle is the injection of a low-frequency modulation on the duty cycle during the constant voltage-phase of charging. The frequency is low so that the frequency response can reveal the battery's low-frequency characteristics. The output, or battery voltage, is then cross-correlated to the input modulation to determine the impulse response. An FFT on this response then provides the battery's frequency characteristics. This analysis should be performed during the time when the battery capacity is not changing, for example during the end of charge.
Another important condition is that the battery's temperature must remain stable, within a few degrees, during the analysis. One other consideration for this analysis is that for multicell packs, charge-balance circuits must be disabled. Due to the low-frequency investigation, the data collection time can be quite long, measured in hours. This “slow” nature also allows the use of the SMBus-enabled batteries to report the voltages for individual cells so that the low-frequency spectral data can be collected on a per-cell basis for multicell packs. Although the analysis time can be long, there are plenty of opportunities during the normal use such as in notebook applications.
This impedance information can then determine various functional battery parameters, such as those in Fig. 2 for Li-ion batteries. These parameters include electrode diffusion and other rate-limiting reactions. It is these rate-limiting characteristics that determine the battery's performance under load.
Digitally Controlled Charging Benefits
Many charger applications use either state machines or microcontrollers to control battery charging. These devices are capable of adjusting the output voltage and current limits to provide the battery's appropriate charging energy. Once the battery-charging system is enabled with the digitally controlled power conversion, the embedded peripherals and computational power can enhance charger management. One common problem faced in charger designs is load sharing. Load sharing is used to maximize power available to charge the battery while maintaining adequate system power. An issue for many cases is how to share power while determining when the battery is full.
A common technique to detect when a battery has a full charge requires the use of a constant current or a current threshold. If the load demands most of the available power, then very little remains available for charging and full-charge detection can be difficult. The adaptive ability of digitally controlled power can be used to combine several indications that the battery is full. For Li-ion batteries, a taper-current threshold most often determines charge termination during the constant-voltage phase of charging. The digital power solution can monitor the situations where the available energy for charging does not meet the requirements and therefore suspends the charge-termination decision.
Charge-Balancing and Digital Power
Multicell battery packs may experience charge-balance issues as the individual cells age. The problem may be indicated when series cells have different potentials. This imbalance is a result of two basic issues. The first is the impedance change in the individual cells as the battery ages. In the Li-ion example, the impedance may double for every 70 cycles. This increase can be greater if the battery is subjected to higher temperatures while at a full-state charge. The other imbalance mechanism is loss of coulombmetric capacity. With Li-ion, this is caused by the loss of lithium due to electrode side reactions.
Charge-balancing attempts to bring all battery-pack series elements to the same state of charge. Although weak-series elements still may have lower coulombmetric capacity, the cell voltages for all cells are brought to the maximum levels to provide the maximum available energy for this situation. In order to reduce the balancing issues due to impedance differences, balancing usually occurs toward the end of charge. For Li-ion, this is when the charging current is the lowest and impedance has the least effect. The digitally controlled solution can use the impedance information to help make balancing decisions. However, when the misbalancing situation is due to impedance mismatch, enabling the balancing function will not be a benefit. Balancing can be performed during the entire charge period if the amount of needed bypass charge for each cell is measured beforehand. This requires a correlation database between open-circuit voltage and state of charge or impedance compensation. Use of either of these methods is made possible due to the programmability and computing power available in a digital-power solution.
Impedance Data and State of Health
Impedance correlation analysis is of special advantage to batteries that are seldom fully discharged. This may be the only means to determine useable run-time or need for replacement outside of a complete discharge. Impedance information is often used to match cells when assembling a battery pack.
In the past, single-frequency impedance measurements (usually at 1 kHz) have been used as a rapid test to identify defective battery cells. The 1-kHz value, which is the value reported in most battery data sheets, can be used in detecting critical defects such as short or open circuits. This method, however, has not been successful in predicting capacity of somewhat degraded but still functioning cells.
Capacity estimation is needed for the quantitative evaluation of a battery's condition, known as the state of health (SOH; SOH = current battery capacity/original battery capacity). Usually, the 1-kHz battery data only provides the resistance of electrodes and electrolytes. On the other hand, a wide-frequency spectrum can identify degradation in active material, which is critical for battery performance.
Once a multifrequency impedance spectrum is available, there are several different methods that can be used to correlate a battery's SOH with the spectral data. In all cases, the correlation function will have a form SOH = f(p1,p2,…pN). Impedance values could be selected at several frequencies directly as correlation parameters p1, p2, etc. However, the choice of two or three frequencies from the entire spectrum would be arbitrary and would omit the remaining correlation information.
Another analysis method involves using all acquired frequency data that are known to have a good correlation to find some derived property, such as the slope of Re(Z) versus Im(Z) in a given frequency range. This method has the advantage of simple implementation. However, it might not make use of all the available data that could be applied to the capacity estimation.
A more generic method that uses all data for SOH estimates is to generate an equivalent circuit model with a matching impedance spectrum. The model parameters can more completely capture the full-dynamic range of the measurement.
For the frequency range from 10 mHz to 1 kHz, a five-parameter model sufficiently captures the spectral content. Fig. 3(a) shows an equivalent circuit valid in this frequency range for most battery types, and Fig. 3(b) shows Li-ion battery spectrum with the frequency response of a fitted model. Such analysis not only reduces data from a large set of data points to only five real parameters, but eliminates the measurement noise. Such a model has to be physically relevant to be able to extract complete information from the spectrum.
Each of the obtained parameters covers certain battery properties. The simple linear correlation with a single parameter in the form SOH=f(p) would therefore only capture correlation of capacity with that property while disregarding the influence of the remaining parameters. On the other hand, multidimensional correlation with all five parameters may exceed the capacity of many smaller microcontrollers.
The usual compromise is to use only two or three circuit model parameters that give the best correlation with target properties. This may have the added benefit of improving correlation because excluding parameters not having good correlation reduces outlier data. For example, serial resistance varies from cell to cell but may not indicate SOH.
Use of several parameters can improve correlation significantly, as can be seen in Fig. 4, which shows battery energy values predicted with one- and three-parameter correlation compared to actual values.
Implementation of impedance-based correlation in UPS or other battery management systems requires the collection of the correlation parameters prior to device operation. However, this collection can be logistically difficult and time consuming.
The correlation information can be stored in a small nonvolatile memory in the battery pack for use in real-time analysis. If such storage is unavailable, then the use of battery normalized parameters and a generic database can be used. Even though correlation error would increase, it would still provide qualitative indication of SOH and identify batteries close to failure.
In addition to correlating circuit model parameters with SOH, an historic analysis of changing parameters can also provide an indication of impending battery failure. For example, a parameter may change according to a fixed linear relationship with time (or with each charge/discharge cycle), but then at some point, the rate of change may follow an entirely different relationship. This indicates the battery may be close to its end of life and needs replacement.
The advantage of this approach in detecting a change in behavior is that no database collection is needed. Only the rate-of-change threshold would need to be preset, and the use of several parameters rather than one would improve battery-change detection robustness.
Digitally controlled power brings new capabilities for battery management. The peripherals and the computational ability of digitally controlled power solutions can make charging and balancing decisions. Charging algorithms can adapt to the available power so that charge termination can be consistent. One of the greatest benefits of digitally controlled power conversion is the battery analytical capability. Using high-resolution ADC, PWM and other peripherals along with computational ability, the battery performance parameters can be determined.
These parameters can be correlated to the SOH of the battery. The end result is that the user can be provided information to aid in determining when the battery needs to be replaced. The additional benefit is the greater level of predictability of a battery's performance that aids in the determination of run time. As digitally controlled power is adopted for battery-powered applications, innovative use of the technology will surely enable better battery management.
“Apparatus and Method for Charging Batteries,” U.S. Patent 4,388,582, June 14, 1983.
Miao, Botao; Zane, Regan; Maksimovic, Dragan; and Button, Robert. “Practical Identification of DC-DC Converter Dynamic Responses and Automated Digital Controller Design,” Digital Power Forum, 2005.
Barsoukov, E., et al. “New Materials for Electrochem Sys.”, 3, (2000) 301.
Rodrigues, S.; Munichandraiah, N.; and Shukla, A.K. “A Review of State-of-Charge Indication of Batteries by Means of AC Impedance Measurements,” Journal of Power Sources, 87 2000 12-20.
Fang, Wei-Chou. “Method of measuring remaining capacity of a storage cell by comparing impedance plot characteristic,” U.S. Patent 5,241,275.
Yoon, C.; Barsukov, Y; and Kim, J.H. “Method of and apparatus for measuring battery capacity by impedance spectrum analysis,” U.S. Patent 6,208,147, 2001.