Mini Project. Classification of 2 Lithium Iron Phosphate Batteries to increase state of charge estimation accuracy.

Transcription

1 Mini Project Classification of 2 Lithium Iron Phosphate Batteries to increase state of charge estimation accuracy. James Moore Registration Number: Wednesday, 01 July 2015 James Moore P a g e 1

2 1 ABSTRACT Two lithium iron phosphate batteries were provided by Moixa Energy for characterisation with an objective of increasing the accuracy of state of charge (SOC) measurements utilizing open circuit voltages (OCVs). The data for this purpose has been gathered by charging and discharging the batteries at various values of current to known values of charge, allowing a settling period to OCV, then introducing a subsequent perturbation and recording the voltages in both states. Batteries were found to be under their rated capacity of 40 Ah with discharge capacities of Ah and Ah at 4 amps constant current. During the process of gathering data, calibrating for the batteries safety limits was straightforward with the previous values. In contrast to Huria, Ludovici et al., 2014, the OCV-SOC curves for these batteries, despite being lithium iron phosphate were not too flat as to prevent this method from working over several current rates if a sufficient resolution of voltage were recorded (2dp or more.) This is however likely to present issues with cheap industrial sensors. The gathered data has the potential to provide a lookup table to determine a state of charge for a given measured voltage and load which would function to give a direct state of charge or as calibration points for a battery management system primarily reliant on ampere counting. It would be suggested to expand on the work completed by to Huria, Ludovici et al., 2014 due to the promise in their method of SOC assessment utilizing Kalman filters for this specific battery chemistry. 2 INTRODUCTION One of the most important characteristics for an electrical storage system, particularly with lithium ion batteries, is having an accurate real time assessment for the state of charge. State of charge is the value in percent of energy that can be extracted from a battery which remains of the total. This is extremely important for both battery engineering and battery end-users. For the former, a more accurate value for state of charge can allow for less designed overhead in capacity making a smaller lighter battery. For a user, they can prevent over charging and under discharging which will decrease uncertainty. This would be particularly useful in vehicle applications due to range anxiety. 2.1 LITHIUM IRON PHOSPHATE Lithium Iron Phosphate batteries (LiFePO 4) or LFP batteries are a type of lithium ion rechargeable battery. They feature a decreased energy density in relation to the lithium cobalt oxide (LiCoO 2) batteries which are typically found in consumer grade electronic devices, but an increased power density and lifetime (Zhang, Yuan et al., 2015). They are also reported to have a relatively low cost of raw materials and are more thermally stable (Hsieh, Liu et al., 2015). There is some discussion in (Huria, Ludovici et al., 2014) regarding the flat curves of OCV which will make SOC determination with in these flat areas unreliable or impossible, this will be observed in the results if present. James Moore P a g e 2

3 2.2 STATE OF CHARGE ASSESSMENT Several methods exist to determine the state of charge of batteries. The most simple is a full battery discharge (the most reliable method) which can take several hours depending on battery size. Another method is ampere (coulomb) counting. A combination of these methods was undertaken in this project to determine safe margins of operation, maximum capacity etc of the provided batteries. Other methods such as impedance spectroscopy or physical property methods (Piller, Perrin et al., 2001) (often used in lead acid batteries to great effect) can be used but are not appropriate in this case due to the chemistry of the batteries and scoping of the project Background Method 1 (Open Circuit Voltage) Open circuit voltage state of charge assessment works by utilizing the relationship of the value of a settled unloaded battery voltage to the state of charge. At a given state of charge after a rest period for the battery to re-equilibrate, the voltage value is determined and the value stored. This can be repeated for a multitude of values to form a large lookup table in which one can find the given state of charge for a battery which has been left to settle. This means that the battery will naturally have to be used in an environment where long settling periods are common although this technique is typically utilized with other techniques in order to adjust or recalibrate. This method is particularly useful in lead acid batteries where there is a linear SOC-OCV relationship leading to accurate results. In lithium ion batteries this can be complicated by varied gradients or flat sections of the SOC-OCV curve. There are also issues with this method with regards to the time required as a rest period and the nature of battery systems utilizing monitoring devices or battery management systems which will draw a small current constantly, resulting in the OCV never being fully attained. These results are also subjective to other variables such as temperature and battery hysteresis Background Method 2 (Ampere Counting) Ampere counting, also known as coulomb counting is a state of charge assessment that works by utilizing the relationship of time and current. With a fixed current over a certain time, the amount of charge delivered can be determined. This method is the most common technique for assessing state of charge. With a relatively simple monitoring system, even variable current can be accounted for with sufficiently low time periods between samples of the current value. This is achieved by plotting the current-time curve and integrating under the curve to find the value of charge integrated (stored.) With a known value of maximum charge and continuous monitoring, an ampere counting system can accurately determine the state of charge by relating current integrated to maximum stored charge. This method works accurately for both charge and discharge due to the nature of the integral naturally adding and subtracting the appropriate values depending on which side of the x-axis of the plot the values lie. There are limitations to this method however; a monitoring system must have a sufficient resolution (small enough time period between samples) or large current spikes over short periods of time can be James Moore P a g e 3

4 missed leading to charge that has been added or removed from the battery but left unaccounted for. Equally, inaccurate measurements of current will lead to significant error. Additionally, batteries are not 100% efficient, so the discharge of a battery will release less charge than was integrated during charging which must be accounted for Chosen Method The state of charge assessment method chosen for these batteries was open circuit voltage based as Moixa Energy had requested this. In order for this method to work, the battery must be left for an adequate amount of time under no current load in order to settle. This could be inappropriate for these batteries if they are intended for solar and overnight charge/discharge they may have a constant load or supply at all times thus the decision was made that a relation must be made from a settled open circuit voltage to various loaded equivalents. This would simulate a state of charge OCV for a battery at a given current load or supply dependant on the loaded voltage. This would then provide a lookup table to determine a state of charge for a given voltage and load which would function to give a direct state of charge or as calibration points for a battery management system primarily reliant on ampere counting. Full discharge tests were also undertaken to determine accurate maximum capacities. 3 EQUIPMENT 3.1 VOLTCRAFT HPS A high power DC bench power supply rated for 600 watts at up to 60 volts and 10 amps. This power supply was used to recharge the batteries at constant current (10 amps) after discharge testing but not during charge testing due to the 10 amp maximum limit. 3.2 VOLTCRAFT PPS A DC bench power supply rated for 300 watts at up to 60 volts and 5 amps. This power supply was only used to supply 18 volts and 1 amp to the relay attached to the power supply or electronic load on the batteries. 3.3 VOLTCRAFT HPS A high power DC bench power supply rated for 900 watts at up to 30 volts and 30 amps. This power supply was used during charge testing due to the adequate amperage rating. 3.4 ELECTRONIC LOAD EA-EL A DC rack mount electronic load rated for 1.2 kilowatts between 0-80 volts and amps. This electronic load was used to discharge batteries as part of the discharge tests at the various constant current rates and after charge testing to discharge the batteries for another test. James Moore P a g e 4

5 3.5 CUSTOM BUILT BATTERY CYCLING RIG A custom built battery rig based off the National Instruments software packages was utilized in order to cycle the batteries fully to determine maximum capacity and the generalized charge-discharge curve trends. This rig connected with the Control Desktop Computer (Section 3.7). 3.6 ARDUINO UNO + RELAY An Arduino Uno microcontroller board was used in tandem with the control desktop computer in order to alternate the connection of electric supply or load to the batteries under test. The Arduino had its own code written in order to change the setting of a transistor, to supply 18 volts at 1 amp to the relay in order to connect relay contacts. The Arduino was interfaced with the desktop computer via serial port over USB, the desktop computer then controlled the Arduino via custom written python code. This python code asks the user for the length of the open circuit relaxation period, the current value chosen and the battery (number) in question. It calculates the appropriate time periods based on these values and the full battery discharge capacity then runs the sequence via the Arduino and relay. The relay utilized for these tests was a TE Connectivity T92S7D CONTROL DESKTOP COMPUTER The control desktop computer utilized Windows 7 with the default python programming language installation provided from Python.org with the following required dependencies: Serial (for communications over serial port.) Time (to keep accurate timing.) CSV (to support optional file output.) Array (to support the usage of programming arrays.) Cmath (to support the automatic calculation of timing.) The control desktop computer was also used to compile and load the Arduino code. Arduino code (C/C++) was compiled, loaded and the Arduino connected by drivers packaged in the Arduino IDE. See: GRAPHTEC MIDI LOGGER GL220 A multichannel data logger capable of recording values between 20 mv and 50 V at samples taken every 10 ms, 20 ms, 50 ms, 100 ms or 1 s. The data logger was connected directly to the control desktop computer and the drivers and software supplied installed. The sample rate was chosen as 100 ms and data recorded directly to the PC hard drive in CSV format. The wire connectors to determine voltage were connected as closely to the batteries as possible, on the load/supply side at the final connector to the battery directly. Accuracy of readings is stated as 0.1 % of filter selected, in this case a filter of 0 50 V giving ±0.05 V. James Moore P a g e 5

6 4 METHODOLOGY 4.1 BATTERY SPECIFICATION The batteries supplied by Moixa Energy were Fullriver lithium iron phosphate battery packs (Model FB- MOA002F) assembled in a 4P 8S configuration for a total of 32 cells each with a nominal voltage of 3.2 volts and a of 10 Ah. This results in a 40 Ah battery with a voltage of 25.6 V. Further characteristics are listed in the table below and the appendix contains the supplied full specification sheet. Table 1 - Fullriver battery specifications Characteristic Specification Remarks Nominal Capacity 40 Ah Under charge/discharge conditions below. Nominal Voltage 24 V Impedance <50 mω AC Impedance 1KHz. 15 A (constant current) charged to 29.2 V, then CV (constant voltage 29.2 V) Charge conditions charge until charge current declines to 0.02 C cut-off or cut off by BMS at 23±3 C. Discharge conditions 15 A (constant current) discharged to 16.0 V cut-off or cut off by BMS at 23±3 C. Operating range 0-50 C. Cycle life 1500 cycles at above conditions 4.2 CONDITIONS SELECTION Conditions for the full cycle maximum discharge testing were set at a constant current of 4 amps as this was the highest stable current possible on that particular testing rig (Section 3.5). The voltage range for this and further testing were selected by constraint. The battery specification was inaccurate and the battery management system embedded within the batteries appeared to cut off when below 24.0 V or above 29.0 V, as such these values were utilized as the range. Current values for the incremental (segmented) SOC-OCV testing were 1, 2.5, 5, 10 and 20 amps respectively, tested under the same voltage range of V. Batteries were tested at standard room temperature and pressure and were not subject to active cooling methods. 4.3 DATA GATHERING / PROCESSING SOFTWARE National Instruments Labview National Instruments Labview (a visual programming development environment) was used with the custom rig by the control desktop computer. A pre-written program was used to run full chargedischarge cycles on the batteries at constant current to determine maximum discharge capacity. James Moore P a g e 6

7 4.3.2 National Instruments DIAdem National Instruments DIAdem (a data visualisation, analysis and processing tool) was utilized to determine the maximum discharge capacity from data recorded in the full charge-discharge tests. 4.4 FULL CYCLE MAXIMUM DISCHARGE CAPACITY DETERMINATION METHODOLOGY Batteries were attached to the custom rig, the pre-written program was started and variables set for the maximum and minimum voltages given in the battery specification sheet. The maximum constant current for that rig was set (4 amps) to move the battery up to the maximum voltage via charging and vice versa. This program was left to run for several cycles on both batteries, a single battery full charge-discharge cycle taking approximately 20 hours at the selected amperage. The data files for these tests were saved in TDMS format by the pre-written program compatible for processing in National Instruments DIAdem. The values of maximum discharge capacity were then calculated by using the integration function of DIAdem and using the Find All Peaks visual basic script, (Section 7.5) on the integrated curve to find subsequent peaks and troughs. Selection of the appropriate peak and trough and correct subtraction of one s value from the other will result in value for maximum discharge capacity in the units utilized for the graph. Subsequent unit conversion is dependent on the previous mentioned units and in this case was converted into amp hours. 4.5 SOC OCV /LOADED EQUIVALENT VOLTAGE DATA GATHERING METHODOLOGY Batteries were attached to the load or supply via the relay, Arduino and desktop PC control assembly after checking that the appropriate test equipment is turned on and the correct cables connected. The voltage values of the connected battery were then checked on the data logger to indicate the correct charge level then for a given test the values for constant current were set on the power supply or load. The data logger was then started recording to a suitably named file name and the python interface code for the Arduino/relay started. This would then prompt the user for the variables required, calculate necessary values and then start the test and move through the various steps. Upon test completion the relay would be left closed. A flowchart of the test procedure can be seen in Figure 1. James Moore P a g e 7

8 Figure 1 - Flow diagram of testing procedure. The variables input by the user were the length of the open circuit relaxation period, the current value chosen and the battery number (which set the maximum discharge capacity.) The flowchart in Figure 2 shows the test logic and loops used in the python code. This shows the loop structure repeating 10 times to move the battery up or down in capacity in ten 10% increments. The disconnect step for 600 seconds shows the open circuit relaxation period chosen for all of these tests. James Moore P a g e 8

9 Figure 2 - Flowchart of python code test logic. 4.6 BATTERY CHARGING AND DISCHARGING NOT UNDER TEST METHODOLOGY Batteries not being tested experimentally were shifted to the appropriate charge state, (fully charged or fully discharged), by constant current of 10 amps to the stated maximum and minimum voltages followed by constant voltage until the value of current hit zero. James Moore P a g e 9

10 5 RESULTS AND DISCUSSION 5.1 CHARGE / DISCHARGE CAPACITY ANALYSIS Initial results and analysis resulted in a modification of testing parameters. A cycle for both batteries was attempted with the voltages stated in the battery specification sheet however the stable range of voltage found by these tests before BMS cut out was found to be between 29.0 and 24.0 volts as stated in section 4.2. Utilizing the method from section 4.4 the values of maximum discharge capacity were found. These can be seen in the table below. Table 2 Full discharge capacity for both batteries determined by charge-discharge cycling. Battery 1 Battery 2 Amp Hours Amp Hours Cycle Cycle Cycle Cycle Cycle Cycle Cycle Avg Avg Std Err Std Err Figure 3 - Battery 1 Charge-Discharge curve at 4 amps constant current. James Moore P a g e 10

11 Figure 4 - Battery 2 Charge-Discharge curve at 4 amps constant current. Figure 3 and Figure 4 show charge-discharge cycles of each battery at a constant current of 4 amps. The curves show that the charging gradient is particularly low in the central section of the charge which will present issues to OCV-SOC assessment with low resolution measurement of voltage. This behaviour is repeated to a lesser extent in the discharge section of the curves. The curves given by each battery match each other quite well despite a slightly lower discharge capacity for battery 2. Further investigation into the flatness of these curves is found in the next section. 5.2 STATE OF CHARGE OPEN CIRCUIT VOLTAGE TESTING ANALYSIS Analysis in this section will take place in two pieces, an analysis of whether the chosen relaxation period was adequate and an assessment of the gathered data for its utility in a state of charge assessment system Relaxation Period Assessment It has been found that the 10 minute (600 second) relaxation period across the different range of test conditions is sufficient in most cases. Observing the relaxation period in both charge and discharge cases at maximum and minimum current rates. Both initial and final relaxations display settling either at completion or nearing completion. These curves can be seen in Figure 5, Figure 6, Figure 8 and Figure 7. James Moore P a g e 11

12 Figure 6 - Final relaxation period for Battery 2 charge tested at 1 amp constant current. Figure 5 - Initial relaxation period for Battery 2 charge tested at 1 amp constant current. James Moore P a g e 12

13 Figure 8 - Final relaxation period for Battery 2 charge tested at 20 amps constant current. Figure 7 - Initial relaxation period for Battery 2 charge tested at 20 amps constant current. James Moore P a g e 13

14 Further equivalent figures for discharge relaxation are un-shown as they show even more favourable characteristics. These are however available in the link given in Appendix 4. This short settling time is a limitation of the study due to the fact that the best settling will occur over a much longer time period, ideally unlimited. It appears that this 600 second time period is an adequate in length, trading-off between the complete settling and the ability to actual gather this data in an operational commercial system. Equally, given the data gathered, an accurate estimate voltage at a shorter or longer time should be trivial either by interpolation or extrapolation. Extrapolation in this case should be valid given the voltage growth/decay should follow an exponential relationship State of charge Open Circuit Voltage Data Utility Assessment Figure 9 - Voltage curve of battery 2 charging at 1 amp constant current with magnified shallow gradient section. Unfortunately both the charge and discharge curves show shallow gradient sections which will present issues in these regions. In the event that the voltage sensor does not have a sufficient resolution, the same value of voltage will match to states of charge between approximately 28.6% and 60%. In the tests performed, the data logger had a sufficient number of decimal places (2) in order to gather useful data. James Moore P a g e 14

15 In Figure 9, the final relaxation voltages from left to right (troughs) in the magnified section are V, V, V, V respectively. This magnified section reflects the typical shallow gradient section found centrally in all plots. Following this trend over the numerous values of constant current it appears that over the value of 26 volts the measuring sensor must have at least 2 decimal place accuracy in order to determine an accurate state of charge. The stated accuracy of the data logger now calls the validity of this data into question as the voltage change between increments is regularly below the error tolerance of ±0.05 volts. It is however likely that the error tolerance is much lower than this as recording data to 2 decimal places is nonsensical if the accuracy only allows rounding to one decimal place. As there has been no clear information on the data logger accuracy, the actual tolerance of the data logger requires more investigation. Figure 10 - Voltage curve of battery 2 charging at 20 amps constant current. As is to be expected, the voltage values vary more extremely with higher current values. This is seen clearly on all segmented charge/discharge figures from low current values with smaller changes in voltage graduating to larger changes with larger current. James Moore P a g e 15

16 Figure 11- Voltage curve of battery 2 discharging at 1 amp constant current. Figure 12 - Voltage curve of battery 2 discharging at 20 amps constant current. James Moore P a g e 16

17 One additional limitation to this method and the data collected is the effect of hysteresis. Hysteresis is defined as the output of a system being dependant as a function of time with current and past inputs. The effect of this is important as all batteries will have hysteresis to some extent and this has not been investigated in this project. The extent of any hysteresis will be important in this battery depending on how the system is utilized. The effect of hysteresis in a solar battery system could be sufficiently different to one doing electrical arbitrage which could result in SOC inaccuracy, or there could be no hysteresis effects at all from either. 6 CONCLUSIONS The selected relaxation time of 600 seconds appears appropriate as it is nearly or is relaxed fully by the time of current interruption. Equally, it is a short time period that could be used in a given system as it is likely possible to rest the battery for 10 minutes. The value of accuracy given by the data logger questions the validity of the data as the voltage increments between increments is regularly below the error tolerance for the device. It is quite likely that the accuracy of the data logger has been miss-stated as an error of on a device reporting 2 decimal place is unlikely as this would only give rounded accuracy to one decimal place. The paper presented by Huria, Ludovici et al., 2014 on SOC estimation on high power lithium iron phosphate (LFP) batteries which challenges the OCV method for determination of SOC on LFP batteries due to flat OCV-SOC curves has been proven as this was observed in this project. It was found that the close voltage values in recorded between increments present an issue unless sufficient data resolution, (2 decimal places can be met.) This is likely to present issues with cheap industrial sensors. The paper also challenges the ampere counting method due to the high power characteristics of LFP batteries leading to the possibility for large integration (and thus SOC) drifts should a large current be used for a small amount of time though this was not investigated in this project. For these reasons it would be a recommended action to not use either method alone, but use ampere counting as the primary SOC measurement with recalibration points provided by the OCV-SOC curves. This is also mentioned as common practise in (Huria, Ludovici et al., 2014.) The limitation of hysteresis is a more complex phenomenon to deal with, however given the success in modelling LFP batteries in (Huria, Ludovici et al., 2014), applying a similar approach with the use of Kalman filtering could result in an extremely accurate gauge of SOC. Not only this, but also excellent error correction from inaccurate estimates with the ability to use cheap industrial sensors without degradation of accuracy. For this reason, it would also be suggested to expand on the work completed by to Huria, Ludovici et al., 2014 due to the promise in their method of SOC assessment utilizing Kalman filters for this specific battery chemistry. However the data gathered in this project should find use in improving the state of charge estimates in an open circuit / load equivalent SOC system or one coupled with coulomb counting. James Moore P a g e 17

18 7 APPENDIX 1 PROGRAMMING CODE / SCRIPTS 7.1 ARDUINO RELAY CONTROL CODE The Arduino control code controls whether the relay is open or closed allowing current to flow or vice versa. This is interacted with over a serial port by the python interface code below. The Arduino code can be found here: PYTHON INTERFACE CODE The python interface code tells the Arduino relay code when to open and close the relay to allow current to flow or vice versa. This code asks for the length of the open circuit relaxation period, the current value chosen and the battery (number) in question. It calculates the appropriate time periods based on these values and the full battery capacity, then runs the sequence via the Arduino and relay. The python interface code can be found here: MATLAB TDMS CSV CONVERSION CODE The Matlab TDMS to CSV code converts the National Instruments TDMS file format into a comma separated value file format. This code can be found here: NATIONAL INSTRUMENTS BATTERY CYCLING CODE The program file for the custom battery cycling rig was coded in National Instruments Labview and can be found here: NATIONAL INSTRUMENTS FIND ALL PEAKS VISUAL BASIC SCRIPT The National Instruments Find All Peaks visual basic script was utilized to analyse the full chargedischarge cycles on the custom test rig. By first integrating the recorded data, then finding a peak to a trough, the value of charge integrated for either charge or discharge can be found. This was used in this manner to determine maximum discharge capacity. This script can be found at the link below: James Moore P a g e 18

19 8 APPENDIX 2 GATHERED DATA The data gathered for this report can be accessed from the below link: 1ZsYm5CdEE5eG5lSURfWWRFNk0&usp=sharing 9 APPENDIX 3 SPECIFICATION SHEETS The specification sheet for the supplied Fullriver batteries can be found at the link below: Specification sheets for the equipment used in these tests can be found at the link below: The specifications for the Graphtec logged can be found at the following URL: 10 APPENDIX 4 FULL REPORT DATA AND ANALYSIS The link below links to the full record of data, produced figures and all associated project data: James Moore P a g e 19

20 11 APPENDIX 5 EQUIPMENT IMAGES James Moore P a g e 20

21 James Moore P a g e 21

22 James Moore P a g e 22

23 James Moore P a g e 23

24 James Moore P a g e 24

25 James Moore P a g e 25

26 James Moore P a g e 26

27 James Moore P a g e 27

28 12 REFERENCES Hsieh, C. T., et al. (2015). Microwave synthesis of copper network onto lithium iron phosphate cathode materials for improved electrochemical performance. Materials Chemistry and Physics Huria, T., G. Ludovici and G. Lutzemberger (2014). State of charge estimation of high power lithium iron phosphate cells. Journal of Power Sources Piller, S., M. Perrin and A. Jossen (2001). Methods for state-of-charge determination and their applications. Journal of Power Sources 96(1) Zhang, B., et al. (2015). Nitrogen-doped-carbon coated lithium iron phosphate cathode material with high performance for lithium-ion batteries. Journal of Alloys and Compounds James Moore P a g e 28