You are on page 1of 8

Modeling Lithium Ion Battery Degradation in Electric

Vehicles
Alan Millner

Massachusetts Institute of Technology Lincoln Laboratory, 244 Wood St., Lexington MA 02420, amillner@ll.mit.edu

Abstract -- A new aging model for Lithium Ion To address this need, the theory of crack propagation in
batteries is proposed based on theoretical models of crack structural materials is applied to the battery electrodes,
propagation. This provides an exponential dependence of giving a nonlinear relationship of cycling stress and
aging on stress such as depth of discharge. A measure of temperature for use in modeling cell damage. This
stress is derived from arbitrary charge and discharge calculation of a damage parameter from cell history is the
histories to include mixed use in vehicles or vehicle to new part of the model, and shows a good match to
grid operations. This aging model is combined with an available cell cycle life data. The resulting cycle life
empirical equivalent circuit model, to provide time and model may then be fitted to a small amount of cell test
state of charge dependent charge and discharge data. This life damage parameter is then applied to the
characteristics at any rate and temperature. This choice empirical equivalent circuit model structure, where it is
of model results in a cycle life prediction with few found to be a good fit to test data. Such a model is
parameters to be fitted to a particular cell. constructed in a MATLAB program capable of reading in
the charge-discharge profile of either lab cell tests, or
actual vehicle driving histories from hybrid electric
INTRODUCTION vehicle simulations, or real vehicle data. This allows a
mixture of complex battery usages to be included in
Rechargeable lithium ion batteries offer high energy and lifetime predictions with limited computational
power density and relative safety, and are therefore the requirements.
battery of choice for many applications from consumer
electronics to electric vehicles. Modeling the cycle and
calendar life of such batteries is needed to set THEORY
expectations of reliable product performance. Models to
predict battery life from limited test data have fallen into The electrochemical modeling of lithium ion batteries and
two types. Models of the detailed chemistry of the cell their degradation with cycling and time is described (3) in
(1), requiring extensive computation facilities, predict terms of several mechanisms. One is the deposition of
some time and temperature effects accurately but do not insoluble lithium precipitates on surfaces in the carbon
agree well with cycle life test data. Cycle life is thought to electrode, removing lithium from the active cell. This can
be related to precipitate formation in cracks propagating progress as new surface is exposed by crack propagation
in the electrode materials of the cells, but linear or exfoliation in the MCMB or graphite structure of the
relationships of cycling parameters such as depth of anode. A second is the separation of anode or cathode
discharge to the resulting electrode damage have not material from the electrical collector metal at either
agreed with highly nonlinear test results. Empirical electrode. Both are essentially physical crack propagation
models fitting test data to equivalent circuit models (2) mechanisms excited by stresses and temperature over
using power law relationships fit a limited range of cycling and time. The second mechanism is less of a
parameters, require extensive testing of each type of cell, factor in modern lithium ion batteries due to the nature of
and have little basis in theory. Further, such models are their fabrication (4) with intimate contact between
limited to fitting simple charge and discharge cycle data, electrode materials. The end effect is a reduction in
and are not easily applied to the mixed cycle behavior of charge capacity and an increase in resistance. Since the
many real applications such as electric or plug in hybrid damage is a single mechanism, it can be represented by a
vehicles. single lumped damage parameter L, varying from 0 to 1
as the cell ages from new to having no capacity left at all.
This allows the model to be split into two parts, the first

978-1-4244-6078-6/10/$26.00 ©2010 IEEE 349

and the second being an equivalent temperature. If the stress level is which are functions of the damage parameter. excited by an external stress added to battery is discharged and recharged over a trip. giving some intuitive value to the parameter L. we introduce the definition of Crack propagation theory offers such a mechanism. a speeds up by a factor of 2 every 10 degrees C. which describes the battery terminal years. First. the resulting loss of capacity is approximately a linear The dependence of degradation due to micro-cycles of function of charge throughput. The basic circuit model would also match the data. Taking this further. In explaining the crack propagation mechanism described in (1). the state of related to the swing in state of charge in a cycle. associated average voltage. Because there is no driving nonlinear representing the normalized deviation of the state of mechanism in prior physical models for this damage charge from its mean over a cycle period. temperature T. we The major problem with the physical models is that they separate the effects of average state of charge (average do not provide a prediction of battery cycle life that is cell voltage. this charge. and we take from the data that the aging process voltage and current as a sum of an open circuit voltage. and temperature. then we get polarization resistance term dependent on state of charge.2. and is widely utilized in battery models (3).0 for the latter effect. To deal with a mixed cycle with an arbitrary SOC. 8). and inversely proportional to t.being the damage calculation accumulated as the value of depth of discharge for every 10 degree C rise in L due to cell history. 7. The time interval defining 350 . To address the two effects. so we need a coefficient linear in approximate doubling of degradation rate for a given normalized throughput charge. In order for the physical depth D with an average state of charge SOCavg must models to address the nonlinear dependence. and an ohmic resistance term. for failure of solid structures. Zhurkov shows a good fit to experimental data for structural failure with his resulting SOCavg = ∫time interval m SOC(t) dt (3) formula for material fracture time t: t=t0 * exp[(U0-gS)/kT] (1) For a cycle in which the SOC is a ramp from 1. This is in fact the Arrhenius relationship.5). the Tafel equation (3).0 down to 0 and back to 1. the cell charge capacity at the C rate at T=25C is approximately 80% of MODELING DAMAGE its original capacity. This thermal dependence has been circuit model of the cell as a function of age L and state of recognized in the literature. then define the normalized deviation as: t0/t = normalized rate = exp[(gS-U0)kT] SOCdev = 2√[3* ∫ time interval m ( SOC(t)-SOCavg)2 dt / =exp(-U0/kT) * exp(gS/kT) (2) ∫total timedt] (4) The resulting rate due to the dominant first term shows Using SOCdev as a fraction of 1. One way random thermal stresses with an activation energy of looking at this is the effect of the overpotential threshold. the Tafel relation) from the stress effects of strongly nonlinearly dependent on depth of discharge. a mechanism cope with the effect of that average state of charge and its is needed with a fundamentally nonlinear nature. the deviation in SOC is 1/2/sqrt(3). It is suggested here that since such reactions may reach equilibrium until new surfaces are By convention. end of life is defined as 80% of original exposed by cracking. this dependence may be due to the capacity remaining at the test discharge condition. it is useful to create the equivalent circuit model so that at the value L=0. These effects can then be represented in The Zhurkov formula indicates that the rate increases an equivalent circuit model as variable series resistances exponentially with the stress level S. and the variable SOCdev types (6. SOCavg = the mean state of charge over the time interval Zhurkov (9) assumes crack propagation to be a thermal of cycling. if we take approach is described by a modified form of the Shepherd the room temperature aging of the battery to be about 10 equation (1. For crack propagation phenomenon instead. Let the time interval m be the mth time the car phenomenon. If we rearrange this to represent a rate of propagation Therefore we invert this value for normalization. . parameter as a function of depth of discharge. This agrees Note that this degradation must be linear over multiple with battery manufacturers’ data (10) showing identical cycles. The details of the second term are difficult. state of charge swing. we aging at modest stresses (calendar life) is highest at high assign a coefficient Kco for the exponential term.0. values for some constants in the Zhurkov equation. we while this dependence is seen in empirical reports of cycle introduce the variable SOCavg representing the average life testing of lithium ion batteries and many other battery state of charge over the cycle. convenience then. but often identified as part of charge SOC.

life. we then have the increment in life parameter here gives an excellent fit over a wide range of possible given by cycles. this can be compared. The fit is excellent.5)/0. let Qnom = the nominal charge For the Peterson A123 data.SOCdev must include both discharge and recharge. The adjusted Peterson data is shown as it would be predicted to be at a 50% average SOC 351 .916. So. particular cylindrical lithium iron phosphate battery intended for PHEV use. and cycles. and Kex matching life at some lower DOD. The would depend on the specific battery. a better comparison for more battery types. This proportional to the concentration of Lithium ions left in has the value of 1. Equating the exponential in the new be complex. and to show the degradation rate is Varta data Battery Cycle Life Models 'Varta Exponential 1000000 model' Peterson A123 data A123 Exponential 100000 model High SOC A123 exponential Cycles model adj to 50% SOC Peterson A123 data adj to 50% SOC 10000 1000 0 20 40 60 80 100 % Depth of Discharge Figure 1. Such simple repeated cycles at varying SOCdev are the only The details of the mechanical strain from dimensional lab data really available for batteries.2 Since the result is nonlinear in SOCdev. For cycles centered on and confuse the analysis of data. To account for this.66E-5 and Kex =0. then returning to full charge. Separation as shown 50% SOC. If the battery cycle is throughput cycles N in the time interval is given by performed starting at full charge and decreasing to a fixed DOD. Using More data over a wider range of parameters will allow Peterson’s data (11) for the A123 ANR26650M1A cell. if we construct change in the carbon anode during intercalation of lithium a nonlinear model matching the data at high and low d ions in relation to cycle parameters are complex. Kco would be set for 100% DOD re-plotted in figure 1 in this fashion. This parameter capacity of the battery and abs(I) = the absolute value of accounts for the effect of voltage averaged over a cycle. SOCdev vs. such a model should yield useful results. the superposition * tcycle/tlife (6) principle may not be applied directly to state of charge variation for battery cycling at arbitrary depths of SOC. the data from (6) and (11) is fitted to battery life data. The combination of the higher degradation due to high SOCdev and lower The factor of 2 arises because we have counted both degradation due to lower SOCavg can obscure the trends charge and discharge currents. log cycle life for Li Ion batteries.0 for the full 100% DOD cycle. Life1 = Kco * N*exp((SOCdev-1) / Kex*Tnabs/Ta) + 0. and is active form. To show this.25)* (1-L) (8) and discharges in the time interval under consideration. Then the degradation is adjusted to account for average state of charge. using the formula proportional to depth of discharge for a single smaller cycle. Then the effective number of matching the Tafel relationship. so Here the value of tlife is the estimated calendar life to the model for such a phenomenon over arbitrary use will 80% capacity.95 were found to be Kco= 3. the values best fitting data for D= 0. We must include the possibility of multiple charges Life2=Life1 *exp(Ksoc*(SOCavg-0. Ksoc = 0.35 to0. battery current at a time t. The value of the resulting constants Kco attests to the good fit of the basic theory behind the form and Kex would be battery specific and are empirically of the model.717. at the end of each cycle. then the larger the N = ∫ time interval m ( abs(I(t))dt/Qnom/2 (5) DOD the higher the SOCavg. the state calculation to the Zhurkov term gives of the material is a function only of its history (it is a state variable) and for simple histories like a series of identical (SOCdev-1) *Tnabs/(Kex *Ta) = gS/kTa (7) cycles of different amplitudes. and must be taken accuracy of the model over a broad range of parameters from test data. However.

there is a lack of such SOC internal resistance. these are suing an Arrhenius relation as predicted by the Zhurkov generally rates below +-5C and temperatures between - model. the entire phenomenon of accelerated degradation at high second low SOC resistance. from 0 ( new) to 1. the work of Liaw(12) is used to establish the form of the result.b.1 0. However. To ratio this to the fixed data in the open literature at present. the cell temperature will be taken as a constant L is the lifetime damage parameter described above. While it is crudely related to the charge capacity modeled here. Tfact = voltage of the battery including a polarization term by: U0/kTnabs^2. and rate.5 profile. For this example cell Rref = 0. We then have the form of the equivalent circuit model. Other data for modern lithium ion batteries tested resistance (limiting capacity). that value must take into account the (12) to give the discharge curves of (10) and be voltage limits for operation and the detailed voltage vs. The first term is mid-SOC resistance. It intervals or vehicle trips is then given by includes a scaling factor Rs to allow different size cells or banks of cells of the same type to be modeled easily form L = Σ ( m=1 to M) Life(m) (10) a basic cell of the same type. compatible with the life parameter L scaling described time curve of the battery. temperature. for this is scaled from the reference model by a factor of Rs *R1. For manufacturer’s data.9 1 hybrid vehicle or utilizing it in a vehicle to grid energy SOC exchange. So.0693. Define the open circuit Where comparing with the Zhurkov formula.5 Model L=0 T=25 1C progressive damage to construct an equivalent circuit Volts Model L=0 T=25 2 10C model that will predict the terminal characteristics of a 1. It is clear that proper resistance R1.5 0.5 Model L=0.2 T=25 1C 0. it is not clear whether Here the parameters a-e are functions of the damage internal heating at high charge rates might account for the parameter L. we here define the dimensionless variable thermal management of the cells is needed to mitigate the resistance scaling factor Rs so that the variable resistance effects of high rate charge and discharge. Discharge Curves A123 Li Ion Cell 4 EQUIVALENT CIRCUIT MODEL 3.4 0. for the cycle m: To complete the equivalent circuit model given the Life(m)=Life2*exp(Tfact*(T-Tnom)*(Tnabs/Ta)) (9) damage parameter L. The manufacturer’s data (10) indicate that Voc = Vnom + SOC*(Vmax-Vnom)/2 the value of Tfact is ln(2)/10 = 0. SOC is adapted from (12) : A rate dependence of battery life due to both charging and R2 = a + b*(SOC)c + d*(1-SOC)e (12) discharging rate is has been documented. for rates within the range discussed. giving a doubling of V = voltage = Voc + (R1 + R2) * I (11) decay rate for each 10 degree C rise in temperature.Finally. 352 . The formula for R2 ( L.0 ( no capacity left). Here it is necessary to restrict the class of cycles to those with rates and temperatures that do not Figure 2.5 Data L=0 T=25 1C Data L=0 T=25 10C 3 The next task is to utilize this theoretical basis for 2. The parameters a. No other rate dependent mechanism is included here. and the last high SOC rates.6 0. including partial cycles and 0 Predicted voltage cutoff multiple cycles.2 0.3 0. Discharge curves for cell model compared with initiate physical processes not included in the model. the battery aging rate is adjusted for temperature modern lithium ion batteries in vehicles. as might be encountered in driving a 0 0. and d for cycle life over a wide range of rates and depths of have dimensions of resistance and scale with the mid- discharge tends to be proprietary. given by the equivalent circuit above: model. where Rs=R1/Rref is the ration of the initial resistance of the cell bank to the This variable L will change over the life of the battery reference cell. This is fitted to the battery selected to reproduce the The damage over the life of the battery for M time values of R2 required at different points in the life L.8 0.1 T=25 1C lithium ion battery for any given charge and discharge 1 Model L=0.011 ohms. giving the total increase in the aging parameter L 20C and +45C (10).7 0. All coefficients have been adapted from left in the battery. model. and may be This avoids needing to re-fit the coefficients to the battery thought of as the fraction of lithium still active in the cell unless its character is very different from the one battery. including any self-heating effects. .

0.71e-3*L3+6. the A123 ANR26650M1A cell in a bank for a PHEV having Ns = 56.6 years without significantly aging the battery. 0. the value of L=0. If the battery were used for V2B at 50% DOD from exponential model.25e-4*L2+4.0645 SOC. T=35C .188e-3*L+5e-4) (13a-e) if ( T < Tnom) (14b) b = Rs*(0. The and the resistance of the bank R1 is scaled by Ns/Np. Np = 11. Capacity loss in simulated cell using terms.8 benign scenario. 50% DOD from full charge. 75% average SOC. This is a state of the art plug-in For a bank of batteries with Ns cells in series and Np cells hybrid electric vehicle battery and has been used in a in parallel. On the other hand.0391*L+5e-3) TempcoR = 1+10*((Tnom-T)/20)2 c = -6.25L + 2. This might represent the use of the battery in a vehicle to grid mode several times a month to level peaks in the electrical power demand of the 0. Limiting depth of discharge for V2B also appears to be a good choice to avoid reducing battery life. of commutes or $1.828*(L+ 2E-3) -0. determine from cell test data compared to the damage model for the parameter L. T=35C 1 batteries for plug-in hybrid vehicle use. and T values allowing terminal voltages and currents to be calculated and charge capacities under any The form of these coefficients is relatively laborious to desired condition to be predicted. the suitability of today’s lithium ion Capacity Loss. 50% swing in full charge.a = Rs*TempcoR*(7.9 temperature of 35C. e = 10. The resulting discharge curves are shown in figure 2. Lower temperature due to a shaded location for the vehicle or a cooler climatic location would represent an even better result. the open circuit voltage is multiplied by Ns large number of retrofit vehicles on the road today. so these deep cycles at high temperature are to be avoided. can be evaluated.69e-5*L +8.95 The model is re-run in figure 3 for the case of 50% swing Charge out (normalized) in SOC.44e-9) This gives a resistive equivalent circuit model for all L. 353 . If this battery cost $5K. Smith (13) points out the need for thermal management of such batteries in parked vehicles in hot climates. about 5000 cycles. Note that the cycle life of 4500 cycles might represent 75 years of 5 cycles a month. T=35C.75 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Cycles Consider the effect of extra V2B cycling in purely dollar Figure 3. with or without V2B. into the following relationships: and the aging ones discharged to 2. the dollar cost per V2B cycle of battery replacement would be $1. If T>= Tnom then TempcoR = TempcoRlo = 1 (14a) Using this model.00 per V2B cycle).0 volts per cell match the capacity curves in the same reference data sheet. tested at C rate. at the 0.85 employer’s facility. shown in figure 4. Keeping battery temperatures below 35C by avoiding peaks in temperature appears feasible and beneficial. a rather 0.2 giving nominally 80% charge capacity can be used as a simpler measure of end The above model is implemented in Matlab and applied to of useful life. The thermal effect on equivalent circuit elements is simplified initial curves are a good match to the vendor’s data sheet. a 100% deep discharge cycle at 45C perhaps up to ten of these per month could easily be with no micro-cycling. swinging from full to half charge.3 TempcoRlo = 1+5*((Tnom-T)/20)2 d = Rs*TempcoRlo*(1. would reduce afforded to capture the $100/month/car of V2B revenue the battery to 80% capacity within 1700 cycles (4. So. so unless the actual equivalent circuit model value R2 is needed to determine capacity at RESULTS different rates. the life to 80% capacity would be SOC.

So it is the deep cycles of In this case the battery was allowed to discharge from driving. Note that with over 200 Figure 6.2 0. time for a 24 mile US06 thousand micro-cycles of life. Figure 5. average SOC = 50%. and the 0.8 0. the 0. This would be only marginally adequate. This 0. Capacity loss in Lithium Iron Phosphate cell. It would use up 20% of the life of 0. maximize gasoline efficiency. overlaid with the micro-cycle 0. micro-cycle SOC swing = CONCLUSIONS 4. clearly not adequate life.3 a typical PHEV.2 by test discharging the simulated cell using the equivalent circuit model. T=35C.2 due to micro-cycles 4.55e-4 of life per trip. Also note that at 45C.86 life would be 2732 trips.9 model will accept those as well. The aging effects of cycling at arbitrary depths of The model is then applied to a simulated vehicle discharge.6 SOC temperature of 45C. if the battery temperature were reduced to 35C 0.94 Charge out (normalized) However.96 effects of driving.84 0. more than adequate life.9 0. 0.88 SOC (which would burn more gasoline).84 battery management is more typical of today’s plug-in hybrids. The and the micro-cycling combined. 0. The basic 1 cycles of vehicular use consist of the long term depletion 0.2 years of daily trips) at 0. When 0. if commuting equated to 25 commute with predictive energy control.86 0.82 0 500 1000 1500 Cycles 0.98 the battery in only 814 trips (2. Figure 5 shows the charge capacity 0.4 0. Note also that the average approximately linear SOC trajectory over the trip to SOC should be kept low to minimize aging effects. 0. that will probably dominate 100% to 20% SOC ( a very deep cycle for such a car) the aging of the battery. T=45C plug-in hybrid (about a 10 mile electric range). average states of charge.8 2 with today’s batteries. and indicates the cycle life Cycles x 10 5 improvements needed. temperatures and commuting profile putting both the long overall discharge combined uses can be evaluated using a new model.7 Figure 4. exponential battery model.3% DOD.5 The simulated battery bank is then exposed to aging due 0. 100% DOD. micro-cycles per day this represents over 20 years of use.98 of the battery each trip.6 1. The plot of state of charge vs. not the micro-cycles.6 0. and at intervals the capacity is measured 0.88 Battery SOC vs Time (Intelligent) 1 0.34%.2 1. State of charge vs. Capacity Loss. Capacity reduction with 100% DOD cycles at a 0.1 0 200 400 600 800 1000 1200 1400 1600 1800 reduction with age in micro-cycles for a temperature of time (seconds) 35C and a discharge rate of 1C. time for a 24 mile commute based on the aggressively driven Charge out (normalized) 0. then the battery 0. and uses an exponential 354 . This simulation uses the approach focuses on crack propagation as the dominant vehicle characteristics of a Toyota Prius with a 5kWh mechanism of degradation.92 actual vehicle performance files are made available. T-35C temperature of 45C is calculated by the model to use 1 2.92 by using a battery thermal management system.94 US06 route repeated 3 times is shown in Figure 6.8 1 1.8 0 0. SOCavg=50%. 0. half this while a controller maintained the SOC along an life would barely be adequate. Such a trip at a maximum Capacity Reduction to L=0.5 years with daily trips. or 7. tested at C rate.96 this rate.4 1.9 discharge limited to a 60% swing between 90% and 30% 0.82 showing that life in PHEVs is barely meeting expectations 0.4 to micro-cycles similar to those often estimated in driving 0.

the exponent for depth of Repelling Force on the Opposite Electrode”. 1 art for future developments. new consistent with crack propagation theory.0 of this aging parameter and its relationship to a life SOCavg = average state of charge over a cycle defined as 80% capacity allows evaluation of different SOCdev = deviation in state of charge in a given cycle m uses and charge control strategies. Ning. respectively Structure with Electrode Particles that Exert a the coefficient of throughput. 151 (10) A1584-A1591. 2008. “Capacity Loss in Rechargeable Lithium Cells DURING Life Cycle Testing: The Importance of Notation: Determining State-of-Charge”. This allows calculation of T = battery temperature in degrees C charge capacity under any desired rate and temperature Ta = battery absolute temperature in degrees K.. 355 . temperature T=25 Vnom = nominal open circuit voltage at SOC = 0. V Svoboda.dependence of ion loss on stresses such as depth of R1 = the equivalent fixed resistance of the battery bank. A. 1121-1125. and Nasar. ThermoAnalytics. Inc. The model to R2 = the variable resistance of the battery. state of charge. average state of charge. Tnabs = Tnom hybrid electric vehicles. means of quantitatively evaluating such advances. using varying deep cycles at + 273 varying average state of charge. Amperes. 2007. The calculation SOC = state of charge. BN. HEVsim Technical Manual. Liaw. White. using micro-cycling at U0 = an activation energy for bond breaking varying average state of charge.851 B2. B. Gozdz et. Ns = number of series cells in a bank 5. R Hwu. J. and if average state of charge is kept low (<60%). 2. workshop average SOC and decrease in lithium ion concentration presentation p. J. 14. positive = charging 4. 81-91. 15 Nov. a function of L this point requires a small number of cell measurements. RE. 2003 L = life aging parameter from 0 to 1.0 7. Electric Np = number of parallel cells in a bank Vehicle Technology. 17 discharge. Ning. J. M. Unnewehr. Ta = T condition with any desired discharge termination voltage. Electrochemical crystal lattice Soc. 0 to 1. chapter “Battery Modeling”. Kex. Rosenkrantz. Ksoc = constants in model equation. US Patent k = the Boltzman constant = 1. Al. These parameters represent the cycle life model of rechargeable Li-ion batteries areas of needed improvements in the state of the battery “. discharge. I = current. 2004. and SOC The resulting life parameter is a measure of ion S = stress applied concentration as a function of cycle life.387.38* 10^-23 J/degC US7. Popov. G and Popov. June 2008. and using actual Vmax = maximum open circuit voltage at full charge and discharge time profiles from simulated vehicle operation. “A generalized guidelines are applied. BN .0 3. t = time in seconds t0 = the period of atomic oscillations in the solid material Beyond this.Y. EVS20: The 20th Life (m) = increment of life parameter L including International Electric Vehicle Symposium and thermal effects for cycle m Exhibition. “Cycle Life Modeling g = a coefficient representing anharmonicity for a given of Lithium Ion Batteries”.5 and T The results show that the battery life can be maintained in Voc = the open circuit voltage of the battery at a given an acceptable range for PHEV use if very deep cycles SOC (<60%DOD) are avoided. age. G. “Self-Organizing Battery Kco. an equivalent circuit model of the battery tcycle = time in seconds of a cycle cell or bank is constructed to provide terminal tlife = total expected shelf life in seconds to 80% capacity characteristics as a function of time.. Dubarry. John Wiley. A123 Systems. and the model provides a Feb 2006. at 25C and 50% SOC and charge or discharge rate. E. Electrochimica Acta 51 (10): 2012-2022. (of Johnson Controls/Varta). S. if temperatures are kept low (<35C). Vehicle to grid operation of a few cycles per month has References: negligible effect compared to driving if these same 1. Power Sources 174. +273 Tfact = coefficient of temperature in thermal aging model The model is then exercised for a current state of the art Tnom = reference battery temperature takes as 25C lithium iron phosphate battery being used in plug-in Tnabs = reference battery temperature K. L. 0 to 1. and the coefficient for average state of charge. Life1 = increment of life parameter in a cycle accounting 1982) for swing in SOC and throughput 6. pp. DOD = depth of discharge. and temperature. Life2 = increment of life parameter accounting for “Plug In Hybrid Batteries”. C.

paper TS04B-4. and Pesaran. 12. Batery Seminar and Exhibit. Power Sources 140 . 10. Florida. A123Systems web site. Smith. Opinions. (1965) “Kinetic Concept of Strength of Solids”. Onori S. 356 . Pwr Sources 195 (2010).. 2009. 2005. and Whitacre. J.N. Ft. conclusions. 11. J. “Lithium –Ion Battery Life Estimation for Plug-In Hybrid Electric Vehicles”. interpretations. Fract. 2385- 2392. 157-161. Jungst RG. Mech. 13.a123systems.com. K. data sheet MD100001-02 for high power lithium ion battery cell ANR26650M1A. 16-19 March 2009. Liaw. 311– 323. J. “Lithium Ion Battery Cell Degradation Resulting from Realistic Vehicle and Vehicle. Vehicle Power and Propulsion Conference 09. 1. Apt J. J. A. www. “Modeling Capacity Fade in LithiumIon Cells”. 26th Internatl. 9. Zhurkov S. 2009. and Doughty DH. Int. T. Nagasubramanian G. Case HL. BY.. 8. This work was sponsored by the United States Government under Air Force contract FA8721-05-C- 0002. Peterson S. Guezennec Y and Rizzoni G. and recommendations are those of the author and not necessarily endorsed by the United States Government. Markel. Lauderdale. Marano V.to Grid Utilization”.. NREL/PR-540-45048.. 6 Oct 2009. “PHEV Battery Trade-Off Study and Standby Thermal Control”.