Impedance Measurement for Advanced Battery Management Systems

8/10/2019 Impedance Measurement for Advanced Battery Management Systems

1/7

EVS27 International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium 1

EVS27 Barcelona, Spain, November 17-20, 2013

Impedance measurement for advanced batterymanagement systems

DA Howey 1, V Yufit 2, PD Mitcheson 3, GJ Offer 4, NP Brandon 2

1 Department of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, [email protected] Earth Science and Engineering Dept., Imperial College London

3 Electrical and Electronic Engineering Dept., Imperial College London4 Mechanical Engineering Dept., Imperial College London

Abstract

We present a fast, low-cost approach to measure battery impedance on-line in a vehicle across a range of

frequencies (1-2000 Hz). Impedance measurement has promise for improving battery management since it

is a very effective non-invasive method of diagnosing the internal state of an electrochemical cell. It is

useful for estimating temperature, ageing, state of charge (SOC) and for fault detection. The aim of this

paper is firstly to explore the usefulness of impedance for estimating SOC, focusing on lithium-ion iron

phosphate (LFP) and nickel manganese cobalt oxide (NMC) cells, and secondly to demonstrate the

performance of a low cost impedance measurement system that uses an existing motor drive similar to thatof an electric vehicle to excite the battery current. We find that measurements made with this system are

accurate to within a few per cent of results from an expensive, bulky commercial system. For SOC

estimation in NMC cells, the charge transfer resistance and SEI layer resistance vary significantly with

SOC. In LFP cells the parameter variation is much less obvious, although the double layer capacitance of

the full pack may be a useful indicator of SOC.

Keywords: impedance, battery management system, electric vehicle, hybrid vehicle

1 IntroductionElectric and hybrid vehicles employ a batterymanagement system (BMS) to manage voltages,currents and temperatures for safety purposes,and to calculate state of charge (SOC), state ofhealth (SOH) and other metrics. With respect toSOC, many systems measure only steady statevalues and use coulomb counting, with anempirical model and lookup table of previouslymeasured parameters to account for changes in

performance at different charge and discharge

rates and with age [1]. More sophisticated

systems use adaptive state estimation to fitdynamic model parameters on-line and deal withsensor noise and uncertainty, e.g. [2] whodescribed a method based upon extended Kalmanfiltering to estimate SOC, power and capacity fade.

Electrochemical impedance spectroscopy (EIS) isa useful tool to investigate the reactions occurringin cells. This involves the application of a smallAC variation in load impedance to a cell atdifferent frequencies to measure the impedance (acomplex number) as a function of frequency. Withaccurate calibration and in combination with othermethods, EIS can be used to gain vital information


2/7


about SOC and SOH of batteries [3]. Impedancemeasurement may also be used to analyze faultsin battery packs [4] for example by comparingthe impedances of multiple cells.

Although EIS is widely used in the laboratory, itis a specialized technique requiring costly, bulkyequipment to achieve good results. Since

batteries typically have very low impedances,accurate measurement is challenging due to thesmall voltage fluctuations for a given currentfluctuation [5]. In this paper we explore a lowcost measurement system for impedance thatcould be used online in a vehicle application,using as much of the existing vehicle electronicsand BMS as possible. Our main initial goal isestimation of SOC from impedance, hence first

in section 2 we examine the expected impedancevariation with SOC for two specific chemistries,

before showing some results from a low costimpedance measurement system in Section 3.

2 Impact of SOC on impedance

2.1 Lithium-ion nickel manganesecobalt oxide (NMC) cells

NMC cells exhibit a clear and well-behavedrelationship between impedance and SOC. A

preliminary investigation, Figure 1, shows themeasured impedance variation from 0.1 Hz to 10kHz for a single 4.8 Ah Kokam prismatic NMCcell (SLPB 11043140H) at various SOC andambient temperature conditions with EISmeasured around OCV in galvanostatic modewith a peak current magnitude of 250 mA using aBiologic VSP potentiostat/FRA system.

0 10 20 30 40 50 60 70 80 90-10

0

10

20

30

40

500.1 Hz

100 Hz

1 Hz

0 % SOC 25 % SOC 100 % SOC

Zreal [m ! ]

- Z i m a g

[ m !

]

10000 Hz

Figure 1: NMC cell impedance at various SOC

Using the equivalent circuit model (ECM) shownin Figure 2, model parameters were fitted tothese data using the ZView fitting software,where L is the series inductance, Rs is the series

resistance, RSEI the solid-electrolyte interphase(SEI) resistance, CPE SEI the constant phaseelement relating to the SEI, RCT the charge transferresistance, CPE DL the constant phase elementrelating to the electrochemical double layer, and W a Warburg impedance modelling diffusion.

Figure 2: Equivalent circuit for fitting OCV EIS data

It was found that the parameters exhibiting thegreatest variation with SOC for the NMCchemistry were the charge transfer resistance andSEI layer resistance as shown in Figure 3, bothexhibiting a monotonic relationship with SOC.This relationship is also found to be consistentfrom one cell of this type to another, see [4]. Theimpact of temperature and SOH on these

parameters must also be accounted for but this isnot discussed here.

0 25 50 75 1000

5

10

15

20

25

R S E I

[ m ! ]

SOC %

0

10

20

30

40

50

60

R C T

[ m ! ]

Figure 3: ECM parameters for NMC cell

2.2 Lithium-ion iron phosphate cellsCompared with NMC cells, lithium-ion iron

phosphate (LFP) cells are more challenging todiagnose. We have previously reported [6] thatSOC determination for LFP cells based on

equivalent circuit fitting of individual cells maynot be conclusive due to a significant spread of thevalues of the fitted parameters from cell to cell.However, individual cell fitted parameters couldnonetheless be useful in estimation of cell healthand for fault detection. It may be more plausible totrack overall pack SOC with one or several fitted

parameters pertaining to the EIS responsemeasured across the entire pack.

In order to investigate this, a small battery packconsisting of four A123 Systems LFP cells

(ANR26650M1A with nanostructured lithium iron


3/7


phosphate cathode and graphite anode) in serieswas assembled. Nominal cell capacity andvoltage were 2.3 Ah and 3.3 V respectively. EISmeasurement was carried out with the samesystem as described previously, operating instack mode, in other words simultaneouslymeasuring the responses of the pack as well asevery cell within it. Each EIS spectrum wasrecorded in galvanostatic mode (i.e. currentcontrol) at (i) open circuit voltage (OCV) with250 mA AC current applied and (ii) under a DC

bias with DC discharge current of 150 mA plussuperimposed 130 mA AC current.

Figure 4 and Figure 5 show the measured EISresponse of the full LFP pack around OCV andunder DC bias respectively at various SOCs and

at constant ambient temperature. It may benoticed that SOC has the most impact on theshape of the EIS curves at frequencies below 50Hz and that there is very little variation withSOC compared to NMC cells. Under DC bias,cells behave non-linearly at very low frequencies(less than 1 Hz) and therefore these diffusionresults have been removed in the DC biased data.

Figure 4: EIS response of 4-cell pack at OCV

Figure 5: EIS response of 4-cell pack under DC bias

Equivalent circuit models (ECMs) were again usedfor fitting parameters, with the same model as

before for the OCV case but this time fitting full pack data. For the DC bias case a modified modelwas used shown in Figure 6, where W the Warburgimpedance is not included in the DC bias ECM forthe reasons previously explained.

Figure 6: ECM for fitting of EIS data with DC bias

The parameters estimated are shown in Figure 7and Figure 8 (values of Rs and L vary only a smallamount with SOC and are therefore not shown).

(a)

(b)

Figure 7: Fitted ECM parameters, full pack at OCV

As can be seen there is much less of a clearrelationship with SOC compared to NMC cells. Itmay also be noticed that there are differences invalues of the parameters under DC bias comparedto OCV. Clearly the impact of DC bias must beaccounted for if impedance is to be used reliably toestimate SOC under load [7]. For consistent resultsthe EIS technique for SOC estimation might bestapplied only when the cells are at rest, aroundOCV, as for example a regular recalibration ofother SOC methods such as coulomb counting.

30 40 50 60 70 80-20

-10

0

10

20

30

3 Hz

0.1 Hz

500 Hz 50 Hz

pack - 100 % SOC pack - 50 % SOC pack - 0 % SOC

Zreal [m ! ]

- Z i m a g

[ m ! ]

5000 Hz

30 40 50 60 70 80-20

-10

0

10

20

30

1 Hz500 Hz

50 Hz

pack - 100 % SOC

pack - 50 % SOC pack - 0 % SOC

Zreal [m ! ]

- Z i m a g

[ m ! ]

5000 Hz

0 20 40 60 80 100

1.6

2.0

2.4

2.8

R S E I

[ m ! ]

SOC %

22

24

26

28

30

R C T

[ m ! ]

0 20 40 60 80 10010

20

30

40

50

C S E I

[ m F ]

SOC %

75

90

105

120

C D L

[ m F ]


4/7


The full pack double layer capacitance shows thevariation with respect to SOC that is closest tomonotonic and therefore may be a reliable single

parameter to use for SOC estimation, althoughthe temperature and SOH dependence of this

parameter must also be accounted for and are notdiscussed here. Alternatively a combination of

parameters could be used. The pack C DL increases significantly below 10% SOC evenunder DC bias, which could be used as a rapidempty detection system. More work is required,and is part of an ongoing research programme bythe authors, to fully understand the impact ofSOC and DC current on impedance, particularlyfor LFP cells, as well as the impact oftemperature which has not been considered hereand is significant [8].

(a)

(b)

Figure 8: Fitted ECM parameters, pack with DC bias

3 In-vehicle measurement ofimpedance

Having shown that impedance could usefullyindicate SOC (particularly for NMC cells), inorder for this to be realized in a battery

management system, a simple and low costhardware and software measurement system isrequired that is suitably accurate. Impedancemeasurement could be incorporated into adrivetrain with an integrated system such as thatillustrated in Figure 9 in relation to an electricvehicle drive. The small signal current perturbationto excite the cells and pack to measure impedancecomes either from variations in the main tractioncurrent due to passive noise on the DC bus fromdriver response or controller response, or from anoptional additional excitation circuit, or from acombination of all of these approaches. Thevoltage and current in the cells and pack ismeasured, amplified and filtered then processeddigitally using a statistical correlation technique todetermine the cells impedance as a function of

frequency. The SOC may be determined from theimpedance frequency response of the packassuming relationships such as those explored inthe preceding section.

Figure 9: Battery impedance measurement in an electricvehicle drivetrain

In this work we focus on the use of a motorcontroller as the excitation source for themeasurement of impedance.

A test rig shown in Figure 10 was constructedcomprising a 30 W DC brushless servo motor withmotor controller directly coupled to a second 30 Welectrical machine to mimic the drivetrain of anelectric vehicle and test the idea of using motorcontroller current excitation to measure batteryimpedance, as explained in our recent paper [9].The approach of using an electrical machine as theexcitation system represents a scaled-down

prototype of a brushless DC drive system. This isrelevant as a demonstrator because of therelationship in the scaling of the relevant

parameters as the drive system is increased in power: in a full size electric vehicle, the battery packs will be larger, but so will the drive currents,and the perturbation current can be larger inabsolute terms as the fraction of perturbation

current to battery capacity can remain the same.

0 20 40 60 80 100

2.4

3.2

4.0

4.8

R S E I

[ m ! ]

SOC %

28

30

32

34

36

38

R C T

[ m ! ]

0 20 40 60 80 100

600

900

1200

1500

1800

C S E I

[ m F ]

SOC %

40

60

80

100

120

140

160

C D L

[ m F ]


5/7


The rig was powered by four A123 Systems LFPcells of the type previously described, connectedin series to form a 12-14 V DC supply. Aflywheel was attached to the motors addingmechanical energy storage to smooth speedfluctuations. A National Instruments dataacquisition system was used for analog anddigital input and output, controlled by acomputer. This produces the current excitationsignal which is fed to the motor controller, andalso measures the voltage across and currentthrough the cells after amplification and filteringthrough a small separate circuit (not shown),which was calibrated as described in [9].

In order to process the current and voltagesignals to extract the impedance from any

arbitrary excitation signal, the cross spectraldensity of the current and voltage and powerspectral density of the current were used incombination with the coherence spectrum as ameans of removing poor quality data.

Figure 10: Motor controller battery test system

Using the test rig, both multi-sine and broadbandnoise current excitation spectra were tested. Theresults were compared to impedancemeasurements made with an electronic load andalso to the Bio-Logic VSP system, using bothmultisine and broadband noise excitation signals,in order to demonstrate the accuracy of this low

cost system and stability of using a motorcontroller as the perturbation source.

4 Results of low cost systemFigure 11 shows the measured impedance (bode

plot) for a single A123 LFP cell comparingmeasurements made using multisine excitationand the commercial EIS system, a low costelectronic load, and the motor rig previouslydescribed. The comparison is good across the fullrange of frequencies investigated (1 Hz to 2 kHz)with RMS uncertainties around 5% in magnitudeand 3 in phase. Using a broadband noise

excitation signal (Figure 12) instead of a multisinesignal achieves even better results, with RMSuncertainties around 2% in impedance magnitudeand 3 in phase. This is because broadband noise isless prone to narrowband interference.

Figure 11: Comparison of results obtained usingmultisine perturbation

Figure 12: Comparison of results obtained using broadband noise perturbation

These encouraging results show that it is possibleto obtain quite accurate impedance measurementson cells using a variety of excitation types andusing a motor controller.

5 ConclusionsThe usefulness of these impedance measurementsfor inferring indirect parameters of the cells or

pack, such as SOC, depends on the exactrelationships between the impedance spectra andthe SOC as discussed in section 2 of this paper.For some cell chemistries, such as NMC, there is a

clear relationship; but for others, particularly LFPthe estimation could be much more challenging.

ElmoPiccolo

+

12-14 VDC

computer

Load~ ~

position feedback

flywheel

Maxon EC45motors

battery pack

100

101

102

103

6

8

10

12

14

16

M a g n i

t u d e ( m O h m )

100

101

102

103

20

15

10

5

0

5

10

15

Freq (Hz)

P h a s e

( d e g

)

Bio Logic VSPMotor Multisine

Elec. Load Multisine

100

101

102

103

6

8

10

12

14

16

M a g n i

t u d e ( m O h m )

100

101

102

103

20

10

0

10

20

Freq (Hz)

P h a s e

( d e g

)

Bio Logic VSPMotor Noise

Elec. Load Noise


6/7


Although there is substantial work still to bedone to demonstrate a complete working BMSthat makes substantial use of impedance, this

paper shows that two critical parts of such asystem are plausible. These are first an accurateimpedance measurement system that can beintegrated into an existing powertrain in anelectric vehicle, and second a clear relationship

between impedance and a desirable parameter toestimate, such as SOC.

AcknowledgmentsThis work was funded by the UK Engineeringand Physical Sciences Research Council, Grant

No. EP/H05037X/1, Advanced batterycondition monitoring in electric and hybrid

vehicles. We are also very grateful for theassistance of Billy Wu at Imperial College whorecorded the EIS data for the NMC cell.

References[1] E. Meissner and G. Richter, Battery Monitoringand Electrical Energy Management Precondition

for future vehicle electric power systems , J. PowerSources, 116(1-2):7998, 2003.

[2] G. Plett, Extended Kalman filtering forbattery management systems of LiPB-based HEVbattery packs Part 1. Background , J. PowerSources, 134(2):252261, 2004.

[3] U. Troltzsch, O. Kanoun, and H. Trankler,Characterizing aging effects of lithium ionbatteries by impedance spectroscopy , Electro-chimica Acta, 51(8-9):16641672, 2006.

[4] G. J. Offer, V. Yufit, D. A. Howey, et al., Fault analysis in battery module design forelectric and hybrid vehicles , J. Power Sources,206:383392, 2012.

[5] L. Lu, X. Han, J. Li et al., A review on the key

issues for lithium-ion battery management inelectric vehicles , J. Power Sources , 226:272288,2013.

[6] Yufit V, Howey DA, Mitcheson PD, OfferGJ, Brandon NP, On-line battery conditionmonitoring using low-cost fast electrochemicalimpedance spectroscopy for automotiveapplications , EIS 2013 Conference, 16-21 June2013, Okinawa, Japan.

[7] J. Jespersen et al., Capacity Measurements of Li-Ion Batteries using AC Impedance

Spectroscopy , EVS24, Norway May 2009.

[8] Srinivasan R, Carkhuff B et al., Instantaneousmeasurement of the internal temperature inlithium-ion rechargeable cells , ElectrochimicaActa 56(17):6198-6204, 2011.

[9] Howey DA, Mitcheson PD, Yufit V et al., On-line measurement of battery impedance , IEEETransactions on Vehicular Transactions,forthcoming (under review).

AuthorsDavid A. Howey PhD (Imperial2010) MA MEng (Cantab. 2002) isUniversity Lecturer in EngineeringScience at the University of Oxford.His research interests are conditionmonitoring and management ofelectric vehicle components including

batteries. He leads projects on fastelectrochemical modelling, model-

based battery management systems, battery thermal management, andmotor degradation.

Vladimir Yufit is a PostdoctoralResearch Associate in the Departmentof Earth Science and Engineering atImperial College London. Heobtained his PhD degree from TelAviv University (Israel) leading

pioneering work on development of

three-dimensional thin film lithium-ion microbatteries. His researchinterests include rechargeable non-aqueous batteries; PEM, alkaline andsolid oxide fuel cells; redox flow

batteries and regenerative fuel cells;super-capacitors and hybrid systems;and advanced characterisationtechniques.

Paul D. Mitcheson received theM.Eng. degree in electrical andelectronic engineering and the Ph.D.degree from Imperial College,London, U.K., in 2001 and 2005,respectively. He is currently SeniorLecturer in the Control and PowerResearch Group, Department ofElectrical & Electronic Engineering,Imperial College London. Hisresearch interests are in powerelectronics across various scales fromenergy harvesting to HVDC includinginductive power transfer interfaces.He also has a stream of work on

battery monitoring.


7/7


Gregory Offer is an EPSRC careeracceleration fellow at ImperialCollege London. He is anelectrochemist in the department ofmechanical engineering working at

the interface between the fundamentalscience and systems engineering offuel cells, batteries andsupercapacitors.

Prof Nigel P. Brandon OBE FREngis Director of the Energy Futures Labat Imperial College London. Hisresearch interests focus on the scienceand engineering of electrochemical

power sources, with a particular focuson batteries and fuel cells.

Documents

Impedance Measurement for Advanced Battery Management Systems