96052 BATTERY ELECTROCHEMICAL NONLINEAR/DYNAMIC SPICE MODEL Michael C. Glass Lockheed Martin Missiles & Space OR7-10 BA50 1111 Lockheed Way Sunnyvale, CA 94089 glass-mike@ im.ssd.lmsc.lockheed.com ABSTRACT: An Integrated Battery Model has been produced which accurately represents DC nonlinear battery behavior together with transient dynamics. The NiH2 battery model begins with a given continuous-function electrochemical math model. The math model for the battery consists of the sum of two electrochemical process DC currents, which are a function of the battery terminal voltage. This paper describes procecures for realizing a voltage-source SPICE model which implements the electrochemical equations using behavioral sources. The model merges the essentially DC non-linear behavior of the electrochemical model, together with the empirical AC dynamic terminal impedance from measured data. Thus the model integrates the short-term linear impedance behavior, with the long-term nonlinear DC resistance behavior. The long-duration non-Faradaic capacitive behavior of the battery is represented by a time constant. Outputs of the model include battery voltage/current, state-of-charge, and charge-current efficiency. The nickel Tafel component is where the Nernst component is and the oxygen Tafel component is The charging (Amp-hour) efficiency is MATH MODEL: The spacecraft battery requires a nonlinear equation formulation, in general, since the observed battery behavior is quite nonlinear in its details. Whereas the solar array model is conventionally terminated in a current source for dynamic reasons, the battery is conventionally terminated with a voltage source output, also for dynamic reasons. This convention has been adopted even in the present case where the battery characteristic is given as a set of voltage-dependent currents. This electrochemical battery model consists of a terminal current which is the sum of two separate processes, the desired Nickel-plate charging current Ini, and the parasitic overcharge (oxygen generation) reaction current I o x . This electrochemical model and parameters were developed by Hafen and Armantrout, and are documented in a separate paper1, 0-7803-3547-3-7116 $4.00 0 1996 IEEE = I n ; -b I O X ~ Ini = Nickel reaction current Ioxy = Oxygen reaction current Ibat 'bat = Battery voltage ncell = Number of series cells SOC = Battery State - of - Charge (in Amp - hours) T = Absolute temperature R, F = Fundamental electrochemical constants 292 Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on December 31, 2009 at 04:54 from IEEE Xplore. Restrictions apply. . . . . . . . . .INTEGRATED . . . . . .BATTERY . . . .MODEL . . . .DYNAMICS ....... SPICE MODEL: NONLINEAR DC COMPONENT A general-purpose SPICE circuit-equivalent for these equations is desired. The circuit-equivalent should be useful for DC orbital analysis, and low- and mid-frequency stability and dynamic analyses. Since the electrochemical equations are DC only, transient and frequency-dependent dynamics must be added somehow. Added dynamic impedances must not disturb the DC accuracy of the model. This model should therefore possess the following characteristics: 4 iFq FLTRsZmI CNF=140F Cp=l.l F I nonlinear DC pulse RLp = 88 mR pulse Lp = 2-5 UH Ls = 0.5 I I I I IT R ~ ~ -14sec * CR ~~ * ~C ~ = 10 msec UH I LP'RLP = 0.1 msec "BAT 1 . DC fidelity to equations 2 . AC transient dynamics Figure 2. Battery Empirical Dynamic Impedance Battery model equations often represent the battery terminal voltage as a function of its terminal current. This form is most convenient for system analysis, since the battery voltage source is a stabilizing node both to the real system, and to the system model itself. In the present case, a voltage-source equivalent is desired, so that impedance elements may be added in series. Because our equations are the form of current as a function of voltage, a transformation of some kind to voltage as a function of current is required. The desired circuit model will have a voltage source output which will be subsequently modified to accept transient impedance elements in series. The initial technique of solving the battery voltage, in a SPICE model, from the battery terminal current, is illustrated in the figure below. The battery model terminal parameters are its voltage Vbat, and its measured terminal (charging) current Ibat. Vbat is calculated as the output of a feedback loop, which integrates the error between the measured Ibat and the total of the calculated target current components Ini and Iox. U Io% ("bat) W f Figure 1. Integration of Vbat(lbat) The loop gain bandwidth product of this current error integration process is given by the following formula, where gbt is the circuit gain parameter and Rbat is the static incremental resistance on the battery I-V curve. For Rbat between 0.02-0.2 a, gbt should be on the order of 1 to maintain 1 Hz accuracy of this loop. ELECTRICAL TRANSIENT DYNAMICS: Pulsed and frequency-response measurements on batteries have revealed a complex impedance behavior. At high frequencies the battery exhibits inductance, with an associated time constant. Over a time span measured in milliseconds, our particular nickel-hydrogen battery exhibits a constant pulsed resistance, together with another component of added resistance having a first-order time lag. And there is a much longer-term resistance or voltage drop component encompassing the remainder of the static or dc Tafel voltage variation with current of the battery. This latter component has been assigned a time constant of 14 seconds based on the available measurements. INTEGRATED BATTERY MODEL These empirically-determined impedances must be integrated into the nonlinear battery model. The nonlinear model, within its transcendental equations, possesses all of the DC resistance of the battery. The dynamic impedance terms must be allowed to operate over defined transient intervals, yielding over to the nonlinear resistance over the longer term, with a specific controlled time constants matching measured behavior. This has been accomplished in the core of the Integrated Battery Model, shown in detail in Figure 3. It is most convenient to cast the transient dynamic elements as purely series subnetworks, which can then be placed in series with a behavioral voltage source. The sum of this unknown voltage source plus the voltage drops across the transient impedance subnetworks must equal the net observed battery terminal voltage. Some of the transient impedance elements are either quite familiar, or easily measured. The high-frequency residual inductance, and the mid-frequency pulsed resistance of the battery are largely determined by the shape and arrangement of the battery's internal metallic conductors. The remaining elements are masked within any simple measurements, and require transient or wide-bandwidth measurements. Simple network synthesis procedures are then used to obtain the desired impedance subnetworks. The resulting inductive and fairly constant resistive components of the dynamic impedance are placed directly in series at the battery model terminals. Only the dc current-carrying resistive parts of the transient impedance network thus obtained are in conflict with the nonlinear battery math model, since the math model represents the entire terminal voltage for dc current. The reason for this goes back to the modeling process itself, in which the math 293 Battery Current I Figure 3 . Nonlinear Dynamic Battery Mode1:Voltage Averaging impedances. If the voltage at v d c is essentially instantaneous as a function of terminal current Ibat, then v d c already contains a term Ibat*C R = Ibat*(Rp + Rs), whenever true physical series resistances having a dc value of R exist. It is then reasonable to remove or isolate the effects of pure resistance by subtracting this term, so that the nonlinear voltage-source component of battery voltage is reduced to Vdc(Ibat) - Ibat*(Rp + Rs). The effects of resistance (Rp, Rs) are recovered by adding the transient subnetworks back explicitly in series. Our complete long-term transient model must include the long time-constant circuit from Figure 2 labeled as Rnf * C,f, model is matched to measurements of the complete battery terminal voltage. It is clear, then, that the voltage source representing the nonlinear battery behavior must then have removed from it the steady-state value of the battery current times the total value of the transient resistances which carry the steady-state current. The core of the nonlinear battery model shows the voltage source for the battery nonlinearity, in series with the selected transient linear impedances. This voltage source is seen to be derived from the previously calculated Vbat(Ibat) dc voltage, with a term subtracted which is Ibat*(Rp + Rs), the steady-state value of the voltage drop through the linear transient Figure 4 . Nonlinear Dynamic Battery Model: Current Averaging 294 Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on December 31, 2009 at 04:54 from IEEE Xplore. Restrictions apply. + I rstrp ml I1 111 I Vac BATTERY OUTPUT feedback I This resulting battery model is easily modified to accomodate Vbat(Ibat) math formulations by making the voltage generator a direct function of Ibat, instead of using the implicit computation or feedback loop used here. One of the critical aspects affecting the static and dynamics of a battery is the nature of the Tafel curves, which govern the I-V characteristic of the battery. The Tafel curves from this battery model are shown in Figure 6 , with state-of-charge as a parameter. The static part of the battery model also includes the nonlinear integrating Nernstian voltage changes that occur as the battery charges and discharges. It can be observed that this battery model has unsymmetrical Tafel voltage about zero current, and unsymmetrical static resistance for charge vs discharge. These features are important to power system dynamics and stability, when transient operations involve transitions across the zero current region. Additionally, stability at or near zero current is strongly influenced by the high Tafel impedance component. a nonlinear element. We can now see that this time constant should apply to the complete nonlinear battery voltage, less the linear transient part which we have now isolated in Figure 3. So then, the voltage source component of the battery output voltage should pass through a time constant with the value of Rnf * Cnf. It can be seen that this time constant governs the time-dependency, of the nonlinear dc output voltage of the battery model, on Ibat. Rnf * Cnf therefore is the lumped voltage-averaging (or current-averaging) time constant of the model. Because the corresponding network in Figure 3 is unloaded, the individual values for Rnf and Cnf are unimportant, as long as their product equals the desired time constant (14 seconds was used here). It is convenient to be able to take the values directly from an empirical model such as the one in Figure 2. We can see that all the transient element values in the SPICE Integrated Battery Model are parameters taken directly from the impedance model of Figure 2. Since Vdc in Figure 3 has a nonlinear dependency on Ibat, the Rnf * Cnf time constant is actually operating on the nonlinear voltage response, rather than on the current itself. If we move this lag circuit to operate directly onto the current Ibat, we obtain a similar circuit which has true battery current averaging in the nonlinear part of the output response. This current-averaging alternative is shown in Figure 4. Because the voltage transient response of this version exhibits the Tafel nonlinearity (Fig. 6 ) explicitly, we believe the basic voltage averaging form of Figure 3 has the greater fidelity. I COMPLETE INTEGRATED BATTERY MODEL The complete Integrated Battery Model includes the behavioral elements which calculate the two current components Ini and Iox. A sequence of behavioral sources has been added to Figure 5 at the upper left. The state of charge (SOC) of the battery is a simple integration of the Ini component of current at the lower left. And the coulombic or Amp-hour efficiency of the battery charge and discharge processes is computed and output at the lower right of the figure. il I B a t t e r y C u r r e n t Amps Figure 6 Battery Tafel Curves The battery voltage dynamics have been simulated and illustrated in the plots below. The first plot shows the fasttransient voltage, including the inductive spike, constant 295 Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on December 31, 2009 at 04:54 from IEEE Xplore. Restrictions apply. V I Battery Voltage Battery Current 50 Wdiv 10 40 Figure 7. Battery Fast Transient Behavior Figure 8. Battery Slow Transient pulsed resistance, and a relatively fast but low-frequency firstorder RC time constant. In the second plot, the longer-term voltage adjustment occurs over a period from 10-30 seconds. The fast transient behavior of this battery model are revealed in the 30-millisecond simulation of Figure 7. The battery current is switched from a substantial discharge to a charge current, through the zero-current region. The early features of the battery voltage transient response are mainly linear in nature. The leading edge features an inductive overshoot transient and recovery governed by Lp, L,, and Rp. The first few milliseconds after the current step exhibit a prompt voltage step controlled by the mid-frequency resistance Rs. For several milliseconds following the current step, the voltage rises an additional increment as the battery exhibits a low-frequency resistance term Rp, with a time constant governed by R p , Rs, and Cp. These parameters were determined from pulsed current measurements on a test battery. The dynamic battery behavior on this time scale is of importance to the performance and stability of regulators, charging circuits, and load switching controls. it owes its origin to the defining exponential or hyperbolic functions. In Figure 8, the full nonlinear resistance of the battery model is exhibited after a first-order time lag chosen to be 14 seconds. This lag is controlled by the fictitious RC circuit Rnf, Cnf whose values are selected to match empirical data. However, the resistance Rnf effects only the time lag in the full nonlinear battery voltage response, not any particular voltage drop component. The delayed change in voltage evident in Figure 8 is the component AIbat*(Rnf-Rs-Rp), where Rnf represents the full nonlinear resistance of the battery math model. The time constant evident in this very low-frequency simulation is of importance in low-bandwidth voltage controls and charge-mode control circuits. ORBITAL SIMULATION PERFORMANCE This battery SPICE model has been successfully integrated into a spacecraft orbital-time-scale energy balance and stability simulation. It has also been useful for low-frequency discrete regulator stability verification. A complete satellite nonlinear electrical power system model for orbital time-scale studies is shown in Figure 9, for a small satellite. One of the objectives of this system model is a demonstration of the utility of a nonlinear analysis program, such as SPICE, for demonstration of system energy balance on an orbital time Over much longer time periods, the battery exhibits a markedly nonlinear I-V characteristic, which we usually think of as a nonlinear static resistance. In our mathematical model, sat-hughesl B"# - I I I ch-control battoxy.dph I 480 L I - I Figure 9. Orbital Energy Balance and Control Stability Model 296 Authorized licensed use limited to: GOVERNMENT COLLEGE OF TECHNOLOGY. Downloaded on December 31, 2009 at 04:54 from IEEE Xplore. Restrictions apply. I scale. This system model now contains complete nonlinear models for the solar array and battery, such as would be expected in a formal energy-balance performance simulation. To facilitate the long simulation time span of this analysis, short time constants have been removed. The rapid progression of this analysis demonstrates the utility of SPICE and similar programs for essentially dc energy balance and system sizing analyses. The simulation results depicted below include dynamic transient effects up to the controller bandwidth of about 1 Hz. The solution includes battery voltage, current, state-of-charge, and charging efficiency for a 100-minute orbital period. This sample analysis, taken for a simple load profile with a sunlight overload, shows the utility and accuracy of SPICE modeling for power system level simulation, when arbitrary and complex component models are used. The speed of execution of this model is many times faster than real time on a variety of personal computers. SPICE DECK4 FOR BATTERY MODEL CIRCUIT NAME: ir.eps.test5g - - START OF DECK .FUNC Max(C,D) (C+D+Abs(C-D))/2 .FUNC Min(C,D) (C+D-AbS(C-D))/2 **** if A > B then 1 else 0 **** .FUNC GT(A,B) Max( (A-B)/Max(Abs(A-B),le-6), 0 ) ****sign(A) .FUNC SG(A) A/Max(Abs(A),le-6) * * * * START ADT SUBCIRCUIT: battoxy.dph (U60) IXY=O.67 W=40 E0=1.334761 * 100=1.42184 * EFF=0.97 * A1=0.01562 * ALF=0.295326 * A2=0.026471 * cAP=33.75 * A3=0.00026778 * EOX=1.557 * KN=1.559547 * RTF=0.0235 * A4=0.10556 * BTA=O.l * TMP=10 * VMN=16 * GBT=10 * NCL=22 * SCI=28.6875 * VBI=28 RRllO 0 10 1K TC=O 0 * START ADT SUBCIRCUIT: DIF (U67) XF114 33 10 180 SUMlS .SUBCKT SUMlS 1 2 3 EADD 3 0 POLY(2) 1 0 2 0 0 1 -1 ROUT 3 0 1MEG .ENDS SUMlS * END ADT SUBCIRCUIT: DIF (U671 * START ADT SUBCIRCUIT: SUM (U66) XF116 60 EO 33 SUM2S .SUBCKT SUM2S 1 2 3 EADD 3 0 POLY(2) 1 0 2 0 0 1 1 ROUT 3 0 lMEG .ENDS SUMZS * END ADT SUBCIRCUIT: SUM (U66) RR89 0 40 lMEG TC=O 0 RR87 0 22 125G TC=O 0 EV93 40 0 22 0 1 CC65 22 0 1 GV92 0 22 0 180 10 HI68 10 0 VHI68 1 VHI68 8 40 0 . 0 EB73 70 0 Value= {(v(8,0)/(22)-(1.334761))/ Figure 10. Orbital Simulation Results CONCLUSIONS An Integrated Battery Model has been produced for SPICE analysis, which combines the behavior of DC battery electrochemical equations, plus a complex set of empirically determined dynamics. A method has been developed to convert I(V) equations to a voltage-source output SPICE model. In addition, a method has been developed to incorporate complex empirical dynamic subnetworks into the voltage output of the battery model, without loss of DC accuracy in the complete model. The dynamic elements and the electrochemical model parameters are all handled as variables, so that the battery SPICE model is a programmable library component at the system level. The battery model has been demonstrated in dynamic simulations, as well as in orbital power system energy-balance calculations. + (0.0235)-(1.559547)*~(170)) RR82 0 70 1K TC=O 0 EB72 170 0 Value= + ~1og(Max(.02/.98.v(140)/((33.75)-v(140))))) RRBl 0 170 1K TC=O 0 RR71 0 140 1K TC=O 0 EB71 140 0 Value= {Min(v(l30),.98*(33.75))} RR69 0 130 1G TC=O 0 EB65 EO 0 Value= + ~(0.67)*~((0.1)*(V(8,0)/(22)-(1.557))/(0.0235))} GV84 0 130 60 0 277.77777777778~ CC63 130 0 1 EVE5 1 0 130 0 1 RR65 0 1 1K TC=O 0 EB76 21 0 Value= + {~((0.295326)*~(70))-exp(((O.295326)-1)*~(70))) RR91 0 21 1K TC=O 0 EB77 60 0 Value= {(1.42184)*~(21)) RR92 0 60 1K TC=O 0 RR93 0 EO 1K TC=O 0 DD62 22 24 M3MCLMI .MODEL M3MCLMI D BV=20K IS=lOf CJO=O N=lOm RS=O DD63 23 22 M3MCLMI W 9 4 24 0 40 W 9 5 23 0 16 RR112 0 14 1K TC=O 0 EB92 14 0 Value= (sg(v(33)*~(60)/max(abs(v(33)). .01)) * END ADT SUBCIRCUIT: battoxy.dph (U60) REFERENCES: (1) D.P. Hafen, "Nickel-Hydrogen Battery Voltage and Current Model", Lockheed-Martin internal report. (2) Douglas P. Hafen, Jon D. Armantrout, "Nickel-Hydrogen Voltage-Efficiency Model", IECEC-96. (3) SPICE Analysis was performed using AWB by Cadence Design Systems, 555 River Oaks Parkway, San Jose, CA. (4) Muhammad Rashid, SPICE for Circuits and Electronics Usine PSPICEB, 2nd ed., Prentice-Hall, 1995. 297