Chapter 13: Making Battery Quick-Test Feasible
When Sanyo, one of the largest battery manufacturers in the world, was asked, “Is it feasible to quick test batteries?” the engineer replied decisively, “No”. He based his conclusion on the difficulty of using a universal test formula that applies to all battery applications, — from wireless communications to mobile computing, and from power tools to forklifts and electric vehicles.
Several universities, research organizations and private companies, including Cadex, are striving to find a workable solution to battery quick testing. Many methods have been tried, and an equal number have failed because they were inaccurate, inconsistent and impractical.
When studying the characteristics relating to battery state-of-health and state-of-charge (SoH and SoC, respectively) some interesting effects can be observed. Unfortunately, these properties are cumbersome and non-linear, and worst of all, the parameters are unique for every battery type. This inherent complexity makes it difficult, if not impossible, to create a formula that works for all batteries.
In spite of these seemingly insurmountable odds, battery quick testing is possible. But the question is asked, “how accurate will it be, and how well will it adapt to continuously changing battery chemistries?” The cost of a commercial quick tester and the ease-of-use are other issues of concern.
Battery Specific Quick Testing
The secret of battery quick testing lies, to a large extent, in understanding how the battery is being loaded. Battery loads vary from short current bursts for a mobile phone using the GSM protocol, to long and fluctuating loads on laptops, and to intermittent heavy loads for power tools.
Because of these differences in loads, a battery for a digital mobile phone should be tested primarily for low impedance to assure a clean delivery of the current bursts, whereas a battery for a notebook should be examined mainly for the bulk in energy reserve. Ultra-low impedance is of less importance here. A battery for a power tool, on the other hand, needs both — low impedance and good power reserve.
Some quick testers simulate the equipment load and observe the voltage signature of the battery under these conditions. The readings are compared with the reference settings, which are stored in the tester. The resulting discrepancies are calculated against the anticipated or ideal settings and displayed as the SoH readings.
The first step in obtaining quick test readings is measuring the battery’s internal resistance, often referred to as impedance. Internal resistance measurements take only a few seconds to complete and provide a reasonably accurate indication of the battery’s condition, especially if a reference reading from a good battery is available for comparison.
Unfortunately, the impedance measurement alone provides only a rough sketch of the battery’s performance. The readings are affected by various battery conditions, which cannot always be controlled. For example, a fully charged battery that has just been removed from the charger shows a higher impedance reading than one that has rested for a few hours after charge. The elevated impedance is due to the increased interfacial resistance present after charging. Allowing the battery to rest for an hour or two will normalize the battery. Temperature also affects the readings. In addition, the chemistry, the number of cells connected in series and the rating of a battery influence the results. Many batteries also contain a protection circuit that further distorts the readings.
Three-Point Quick Test
The three-point quick test uses internal battery impedance as a basis and adds the battery voltage under charge and discharge to the equation. The readings are evaluated and compared with reference settings stored in the tester. Let’s explore each of these fundamentals closer to see what it entails:
Internal resistance — To measure the impedance, a battery must be at least 50 percent charged. An empty or nearly empty battery exhibits a high internal resistance. As the battery reaches 50 percent SoC, the resistance drops, then increases again towards full discharge or full charge. Figure 13-1 shows the typical internal resistance curve of a NiMH as a function of charge. Note the decrease of impedance after the battery has rested for a while. To obtain accurate results, allow the battery to rest after discharge and charge. [13.1]
Charge Voltage — During charge, the voltage of a battery must follow a narrow predetermined path relating to time. Anomalies such as too high and too low voltages are identified. For example, a fast initial rise reveals that the battery may be fully charged. If the voltage overshoots, the battery may be ‘soft’. This condition often arises when one or more cells have developed dry spots. A frozen battery exhibits a similar effect. If, on the other hand, the voltage does not increase in the allotted time and remains constant, an electrical short is suspected.
Discharge Voltage — When applying a discharge, the voltage drops slightly, and then stabilizes for most of the period in which the energy is drawn. As the battery reaches the end-of-discharge point, the voltage drops rapidly. Observing the initial voltage drop and measuring the voltage delta during the flat part of the discharge curve provides some information as to the SoC. However, each battery type behaves differently and an accurate prediction is not easy. NiCd batteries that have a long flat voltage during most of the discharge period are more difficult to predict using this method than chemistries which exhibit a steady voltage drop under load.
Unfortunately, the battery’s SoC affects the three-point quick test. Even within a charge range of 50 to 90 percent, fluctuations in the test results cannot be avoided. Internal resistance readings further influence the final outcome. If used as a linear correlation with capacity, internal resistance measurements can be highly unreliable, especially with NiCd and NiMH batteries. Figure 13-2 compares the accuracy of six batteries when tested with the three-point quick test. To establish the true capacity, each battery was analyzed by applying a full charge/discharge/charge cycle.
Often referred to as the ‘Feel Good Battery Tester’ because of overly optimistic readings, the three-point quick test method fails to provide the accuracy and repeat-ability that serious battery users demand. [13.2]
The impression of casual battery users that this method is “better than nothing” will not stand up to the requirements of critical industries such as biomedical, law enforcement, emergency response, aviation and defense. Because of relatively low cost, the three-point tester finds a strong niche in the consumer market where a wrong reading is simply a nuisance and does not threaten human safety. Satisfactory readings are achieved in the mobile phone market where batteries are similar in format. It should be noted that the three-point quick test method provides better results than merely measuring the battery’s internal resistance or voltage.
The Evolving Battery
The Li-ion battery has not yet matured. Chemical compositions change as often as once every six months. According to Moli Energy, a large manufacturer of Li-ion batteries, the chemical composition of Li-based batteries changes every six months. New chemicals are discovered that provide better load characteristics, higher capacities and longer storage life. Although beneficial to consumers, these improvements wreak havoc with battery testing equipment that base quick test algorithms on fixed parameters. Why do these changes in battery composition affect the results of a quick tester?
The early Li-ion batteries, notably the coke-based variety, exhibited a gradual drop of voltage during discharge. With newer graphite-based Li-ion batteries, flatter voltage signatures are achieved. Such batteries provide a more stable voltage during most of the discharge cycle. The rapid voltage drop only occurs towards the end of discharge.
A ‘hardwired’ tester looks for an anticipated voltage drop and estimates the SoH according to fixed references. If the voltage-drop changes due to improved battery technology, erroneous readings will result.
Diverse metals used in the positive electrode also alter the open terminal voltage. Manganese, also referred to as spinel, has a slightly higher terminal voltage compared to the more traditional cobalt. In addition, spinel ages differently from cobalt. Although both cobalt and spinel systems belong to the Li-ion family, differences in readings can be expected when the batteries are quick tested side-by-side.
The Li-ion polymer has a dissimilar composition to the Li-ion and responds in a different way when tested. Instruments capable of checking Li-ion batteries may not provide reliable readings when quick testing Li-ion polymer batteries.
The Cadex Quicktest™ Method
A battery quick text must be capable of adapting to new chemical combinations as introduced from time to time. Cadex solves this by using a self-learning fuzzy logic algorithm. Used to measure analog variances in an assortment of applications, fuzzy logic is known to the industry as a universal approximator. Along with unique learning capabilities, this system can adapt to new trends. Similar to a student adapting to the complexity of a curriculum, the system learns with each battery tested. The more batteries that are serviced, the higher the accuracy becomes.
Cadex Quicktest™ is built on the new Cadex 7000 Series battery analyzer platform. This system features interchangeable battery adapters that contain the battery configuration codes (C-codes). When installed, the adapter sets the analyzer to the correct battery parameters (chemistry, voltage rating, etc.).
To enable quick testing, the battery adapters must also contain the matrix settings for the serviced battery. While matrices for the most common batteries are included when acquiring the adapter, the user is asked to enter the information on those adapters that have not yet been prepared for quick testing. This can be done in the field by ‘scanning’ the working battery.
The ‘Learn’ program of the Cadex 7000 Series battery analyzer performs this task by applying charge-discharge-charge activities on the test battery. Similar to downloading a program into a PC, the information derived from the battery sets the matrices and prepares the Cadex Quicktest™ function. The ‘Learn’ program completes its cycle within approximately four hours. One learning cycle is the minimal requirement to enable the Cadex Quicktest™ function.
With only one battery learned or scanned, the confidence level is ‘marginal’. Running additional batteries through the learning program will fill the matrix registers and the confidence level will increase to ‘good’ or ‘excellent’. Like a bridge that needs several pillars for proper support, the most accurate quick test results are achieved by scanning individual batteries that have SoH readings of around 100, 80 and 60 percent. The confidence level attained for a given battery adapter is indicated on the LCD panel of the analyzer.
The Cadex Quicktest™ can be performed with charge levels between 20 and 90 percent. Within this range, different charge levels do not affect the readings. If the battery is insufficiently charged, or has too high a charge, a message appears and the analyzer automatically applies the appropriate charge or discharge to bring the battery within testing range. Charging or discharging a battery immediately prior to taking the reading does not affect the Cadex Quicktest™ results.
The reader may ask whether the Cadex Quicktest™ system can also learn incorrectly. No — once the learning cycles have been completed for a given battery, the matrix settings are firm and resilient. Testing bad batteries will not affect the setting.
Spoilage is only possible if a number of bad batteries are purposely put through the ‘Learn’ program in an attempt to alter the existing matrix. Such would be the case when scanning a batch of batteries that have not been properly formatted, have been in prolonged storage, or are of poor quality. To protect the existing matrix from spoilage when adding learning cycles, the system checks each new vector reading for its integrity before accepting the information as a valid reference. Learned readings obtained from defective batteries are rejected.
If a battery adapter has lost its integrity as part of ‘bad learning’, the matrix setting can be erased and re-taught. As an alternative, Cadex will make recommended matrices available on the Internet. Users may also want to exchange learned matrix information with each other. Copying battery adapters by inserting a recognized adapter into the analyzer will achieve this. Another method is ‘Webcasting’ the matrices over the Internet.
How does the Cadex Quicktest work?
The first stage of the Cadex Quicktest™analysis uses a waveform to gather battery information under certain stresses, establishing probability levels for the given battery. Since there are many battery types with several interacting variables, a set of rules is applied to further evaluate the data. The results are averaged and an estimated battery capacity is predicted. The initial inference to categorize the batteries is computed from a set of specialized shapes called membership functions. These membership functions are unique to every battery model and are developed using a specialized trend-learning algorithm.
The raw data, consisting of three or more items, flows through the input layer. Vectors leading from the input layer are weighted and the derived values are passed through a function in the hidden layer. Another vector set channels the information to the output. [13.3]
The weights are highly significant and function as the learning facility of the network. A run would proceed with a certain set of weights. If the result is off by a certain range, the weights are changed and the process is repeated until a certain number of iterations have passed or the algorithm produces the correct output.
The Cadex Quicktest™ requires less time than most other methods. While current quick test systems, such as those used in defense applications, need hundreds of learning cycles and run on large computers, the Cadex method requires minimal experience and can be performed on relatively simple hardware. Typically less than five learning cycles are necessary to achieve robust, model-specific solution sets, also known as matrices. This massive reduction in time is the result of a new self-learning algorithm that acquires numerous measures of the battery’s characteristics. The algorithm uses a unique decision-making formula that determines the best solution set for each battery model.
Of course, artificial intelligence is a complicated subject, and is beyond the scope of this book. With respect to complexity, Dr. Lofti Zadeh spoke these famous words: “As complexity rises, precise statements lose meaning and meaningful statements lose precision.”
Battery quick testing has raised the interest of manufacturers and users alike. The race is on to provide a product that is accurate, easy to use and cost effective. The true winner may not be an individual or organization that amasses the largest number of patents, but a company that can offer a product that is cost effective and truly works.
Battery Testing and the Internet
Increasingly, the Internet plays a pivotal role in battery testing. The ability to send all battery test results to a central global database is an exciting prospect. With this information on hand, battery manufacturers would be able to perform battery analysis based on battery type, geographic area and user pattern. Field failures could be identified quickly and appropriate corrections implemented.
Another application for the Internet is establishing a global database for all major battery types, complete with matrix settings. With compatible systems, users would be able to select and download battery information from a central database. Batteryshop™, a software product offered by Cadex, provides such a service. The database lists all common batteries, complete with battery specifications and matrix information. Point and click technology programs the battery analyzer to the correct battery parameters.
Collaborating with battery manufacturers enables Cadex to create the most accurate vector settings. Manufacturers welcome such a system because it reduces beta testing and puts the manufacturer in closer contact with the battery user. The aim is to reduce warranty returns and increase customer satisfaction.
Another powerful feature of the Internet is downloading new software for hardware upgrades. Since battery quick testing is still in its infancy, improved software will be made available in the future that allows upgrading existing equipment with the latest developments.
Electrochemical Impedance Spectroscopy
Electrochemical Impedance Spectroscopy (EIS) has been used for a number of years to test the SoH and SoC of industrial batteries. EIS is well suited for observing reactions in the kinetics of electrodes and batteries. Changes in impedance readings hint at minute intrusion of corrosion, which can be evaluated with the EIS methods. Impedance studies using the EIS technology have been carried out on lead acid, NiCd, NiMH, Li-ion and other chemistries. EIS test methods are also used to examine the cells on larger stationary batteries.
In its simplest manifestation, measurements of internal battery resistance can be taken by applying a load to a battery and observing the current-voltage characteristics. A secondary load of higher current is applied, again noting the voltage and current. The current and voltage relationship of the two loads can be utilized to provide the internal resistance using Ohm’s Law.
Rather than applying two load levels, an AC signal is injected. This AC voltage floats as a ripple on top of the battery DC voltage and charges and discharges the battery alternatively. The AC frequency varies from a low 100mHz to about 5kHz. 100mHz is a very low frequency that takes 10 seconds to complete a full cycle. In comparison, 5kHz completes 5000 cycles in one second. At about 1000Hz, the load behaves more like a DC resistance because the chemistry cannot follow the rapid changes between charge and discharge pulses. The information about electrolyte mass transport is ascertained at lower frequencies.
Additional information regarding the battery’s condition can be obtained by applying various frequencies. One can envision going through different layers of the battery and examining each level. Similar to tuning the dial on a broadcast radio, in which individual stations offer different types of music, so too does the battery provide different information of the internal processes. The EIS is an effective technique to analyze the mechanisms of interfacial structure and to observe the change in the formation when cycling the battery as part of everyday use.
When applying a sine wave to a battery, a phase shift between voltage and current occurs. The reactive load of the battery causes this phenomenon. The overall battery resistance consists of three resistance types: pure resistance, inductance and capacitance. Capacitance is responsible for the capacitor effect; and the inductance is accountable for the so-called magnetic field, or coil effect. The voltage on a capacitor lags behind the current. This process is reversed on a coil and the current lags behind the voltage. The level of phase shift that occurs when applying a current through a reactive load is used to provide information as to the battery’s condition.
One of the difficulties with the EIS method is interpreting the information. It is one thing to amass a large amount of data, and another to make practical use of it. Although the derived information reflects aging and other deficiencies, the readings are not universal and do not apply in the same way to all battery makes and types. Rather, each battery type generates its own set of signatures. Without a library of well-defined reference readings with which to compare, the EIS method has little meaning.
Modern technology can help. The vector settings of a given battery type can be stored in the test instrument and translated into meaningful readings by software. The readings can further be analyzed by coupling impedance spectroscopy with a fuzzy neuro-adaptive algorithm.
Electrochemical Impedance Spectroscopy is commonly used to research batteries in a lab environment. Best results are obtained on a single cell. EIS is also used in aviation and in-flight analysis of satellite batteries. Closer to earth, the EIS method examines stationary batteries for grid corrosion and water loss. Further refined, the EIS technology has the potential for wider applications, such as testing portable batteries. EIS may one day test batteries in a matter of seconds and achieve higher accuracy than current methods.