applied

sciences
Article
Effects of Vibration on the Electrical Performance of
Lithium-Ion Cells Based on Mathematical Statistics
Lijun Zhang 1, * ID
, Zhansheng Ning 1 , Hui Peng 1 , Zhongqiang Mu 1 and Changyan Sun 2
1 National Center for Materials Service Safety, University of Science and Technology Beijing, Beijing 100083,
China; ustb41140007@163.com (Z.N.); peng111hui@126.com (H.P.); s20161186@xs.ustb.edu.cn (Z.M.)
2 School of Chemistry and Biological Engineering, University of Science and Technology Beijing,
Beijing 100083, China; cysun@ustb.edu.cn
* Correspondence: ljzhang@ustb.edu.cn; Tel.: +86-10-6232-1017

Received: 20 June 2017; Accepted: 5 August 2017; Published: 7 August 2017

Abstract: Lithium-ion batteries are increasingly used in mobile applications where mechanical
vibrations and shocks are a constant companion. There is evidence both in the academic and
industrial communities to suggest that the electrical performance and mechanical properties of the
lithium-ion cells of an electric vehicle (EV) are affected by the road-induced vibration. However,
only a few studies related to the effects of vibration on the degradation of electrical performance of
lithium-ion batteries have been approached. Therefore, this paper aimed to investigate the effects
of vibration on the DC resistance, 1C capacity and consistency of NCR18650BE lithium-ion cells.
Based on mathematical statistics, the method changes of the DC resistance and the capacity of the
cells both before and after the test were analyzed with a large sample size. The results identified that
a significant increase in DC resistance was observed as a result of vibration at the 95% confidence
level, while typically a reduction in 1C capacity was also noted. In addition, based on a multi-feature
quantity, a clustering algorithm was adopted to analyze the effect of vibration on cell consistency; the
results show that the cell consistency had deteriorated after the vibration test.

Keywords: lithium-ion battery; vibration load; electrical degradation; mathematical statistic;

consistency analysis

1. Introduction
With the successful promotion of Tesla, electric vehicles have shown a rapid development trend
in recent years. As the key technology of electric vehicles, research on the energy storage devices of
lithium-ion batteries has become a state-of-the art research field [1]. Some researchers are focused
on preparing new kinds of lithium-ion batteries, while others are studying the influence factors of
lithium-ion batteries to improve their service life, mechanical durability and reliability. Compared with
conventional factors such as the C-rate, the depth of discharge (DOD) and temperature, the vibration
has not attracted widespread attention. Vibrations resulting from road roughness, acceleration
inertia and sudden collision will seriously affect the mechanical properties and electrical performance
of batteries.
The fatigue failure caused by vibration is a common problem in the research area in electrical
power systems. The electrode material of lead-acid batteries [2,3] and the mechanical structure of
proton exchange membrane fuel cells (PEMFC) [4] were deformed and cracked under the condition of
the certain vibration load. In terms of electrical performance, it has been found that the voltage and
power of the PEMFC did not obviously change after vibration and shock tests [5].
In the past 5 years, the influence of the mechanical force on lithium-ion batteries has attracted
more and more attention from researchers. Some researchers have studied how to utilize vibration to
achieve a health diagnosis [6] or energy harvesting [7]; while, at the same time, the battery array is

Appl. Sci. 2017, 7, 802; doi:10.3390/app7080802 www.mdpi.com/journal/applsci

Appl. Sci. 2017, 7, 802 2 of 21

configured to prevent the effects of vibration on the electric vehicle [8]. Via the static analysis of the
battery pack level, the stress distribution of mechanical structure of the battery pack has also been
considered to determine deformation induced by vibration load. In Ref. [9] the vibration durability of
the battery tray was studied through experimentation and simulation techniques. The results showed
that the load spectrum simulation experiment could reflect the road-induced vibration well. At the
same time, it was also found that the vibration in the Z-axis direction was the most dominant excitation
when compared to other axes. Moreover, the finite element model of the battery tray [10–12] was
established to analyze the intrinsic mode and the characteristics of the responses caused by the random
vibration. Based on the above, researchers have set out the design of the mechanical durability and
electrical durability for the battery packaging structure [13,14].
In the past three years, research concerning the effects of vibration on single battery cells has
been carried out from different aspects and provided different sorts of insight into the effects of the
vibration on the electrical performance and the internal mechanical structure. At present, the effects
of the vibration on the electrical performance have not yet been met by consistent conclusions. It is
worth noting that the conclusions within Refs. [15,16] were opposite to each other. It is reported in
Ref. [15] that the LiCoO2 /mesocarbon microbeads (MCMB) battery displayed an increase by 3.77%
in the ohmic resistance and displayed a reduction by 1.04% in the 1C capacity after vibration testing,
while it is reported in Ref. [16] that the internal resistance of the lithium-ion battery was significantly
decreased and the capacity was increased slightly after vibration testing. It has also been noticed that,
within Refs. [15,16], conclusions are restricted due to the limited sample size employed, since each
test condition was evaluated only by two single cells. Concerning this problem, some scholars have
conducted further research. In Ref. [17], the method by which these mechanical loads affect lithium-ion
cells was researched, and the results showed that in 18,650 cells stressed in the Z-direction the mandrel
dispatched itself and struck against internal components, which were observed with a micro-computer
tomograph, but the electrical performance exhibited no degradation. However, within Ref. [17], the
3 h load profile was carried out ten times with an inspection, so that battery resting time was not
enough, resulting in the measurement uncertainty. In addition, each test condition was evaluated
only by two single cells, and accidental errors also cannot be neglected, such as a sudden increase in
battery resistance for the long-term vibrational test. In summary, while studies above have identified
the effects of the vibration on the mechanical structure inside the lithium-ion cells, it is ambiguous
whether the vibration had a significant effect on the electrical performance of lithium-ion cells. With
regard to the problem of whether the deformation of the mechanical structure inside cells would lead
to the degradation of electrical performance, research in Ref. [18] noted that the capacity fading of
LiFePO4 batteries was caused by the formation of cracks inside the electrode material. Therefore, it
is reasonable to suspect that the electrical performance of cells would be affected by the structural
deformation and cracks inside cells induced by the vibration loading.
Hooper, J. and Marco, J. from Warwick University, dedicated to studying the vibration durability
of lithium-ion batteries, have performed a great deal of related research works and efforts in recent
years. In Ref. [19], the authors emphasized that studies on the vibration characteristics of the battery
pack should attract more attention; in Ref. [20], the cells natural frequency was proposed by authors for
the first time to quantify the mechanical characteristics of cells; in Refs. [21–23], 18,650 lithium-ion cells
of different materials were investigated with different load profiles; and in Ref. [24], the authors defined
a process to devise random power spectral density (PSD) profiles utilizing vibration measurements
from three contemporary EVs, for undertaking vibration durability evaluations of underfloor mounted
rechargeable energy storage systems (RESS). In brief, two scholars have opened up a wide range of
the research space for this issue. It is reported that, within Ref. [21], these nickel manganese cobalt
oxide (NMC) cells displayed a significant increase in ohmic resistance within the range of 25.82 to
38.09% after vibration testing, and a reduction in the measured 1C energy capacity was observed in
some samples, with 12.22% being the greatest reduction observed. Due to the limited sample size
employed, therefore, authors performed further research within Ref. [23], and this study identified
Appl. Sci. 2017, 7, 802 3 of 21

that both the electrical performance and the mechanical properties of the nickel cobalt aluminum
oxide (NCA) lithium-ion cells were relatively unaffected when exposed to the vibration energy that
is commensurate with a typical EV life. However, the study also noted that a similar reduction in
energy capacity and increase in DC resistance was witnessed within the reference samples, which
indicated that the change in these electrical attributes was a function of other environmental conditions.
Because the total duration between pre and post-test characterization was 3 months, the effect between
the cell ageing and the vibration cannot be fully quantified. The authors also suggested that a further
set of repeat tests should be conducted to increase the size of the data set, indicating that a greater
sample size is necessary to increase confidence of findings.
From the review above, it can be seen that there is a problem when the sample size is too small;
results are greatly affected by some accidental errors such as the cell-to-cell variation, the environmental
factors and the measurement uncertainty. Consequently, it is difficult to identify whether the vibration
or the accidental error leads to the change in the electrical performance of cells. Therefore, it is necessary
to evaluate the change in the electrical performance of cells overall with a larger sample size. In this
paper, 32 NCR18650BE lithium-ion cells are investigated with the load profile according to the standard
Society of Automotive Engineers (SAE) J2380, and changes in the electrical performance of the samples
before and after the test are analyzed by the mathematical statistic method. In addition, this paper also
studies the effects of vibration on the cell consistency, based on the hierarchical clustering method.
This paper is structured as follows: Section 2 provides a detailed description of the experimental
procedure, including the electrical performance testing method and the design of the vibration test
profile. Results are presented in Section 3, and an assessment of the 1C capacity degradation is included
as well as changes in DC resistance of the cell. Section 4 analyses the consistency of the cell before and
after test. Conclusions are presented in Section 5.

2. Experimental Method

2.1. Determination of Sample Size

The determination of the sample size is critical to the accuracy of the test result. In general, the
smaller the number of samples, the greater the error of the statistics, but the increase in the sample size
is subject to many restrictions. In order to achieve the purpose of testing and minimize the error, the
similar sample standard deviation σ and the acceptable estimation error E were known at a certain
confidence level 1 − α in this paper. An equation of the sample size calculated could be derived as the
following:
( z )2 σ 2
n = a/22 . (1)
E
According to the prior data, which was tested by the same measurement method and the
environment using the hybrid pulse power characterization (HPPC) test with an ambient temperature
of 25 ◦ C ± 3 ◦ C around the kind of cells, the standard deviation of the resistance of the same batch
battery is about 0.8 mΩ [25]. An estimation error of less than 1% is assumed. From the central limit
theorem, the mean value of samples will obey the normal distribution for a sufficiently large sample
size (n ≥ 30). Therefore, the value of the sample estimation error E is considered to be 0.28 mΩ, and
the calculation result of n ≥ 31.36 was obtained at a 95% confidence level; therefore, 32 lithium-ion
cells are used in this experiment.

2.2. Test Electrical Performance

2.2.1. The Program Design of Cell Sample Test

When accomplishing the electrical performance test of cells, internal resistance measurements
are, in particular, very sensitive to electronic instrumentation and mechanical preparations; therefore,
some precautions should be taken to produce consistent measurements throughout testing. A testing
Appl. Sci. 2017, 7, 802 4 of 21
Appl. Sci. 2017, 7, 802 4 of 21
Appl. Sci. 2017, 7, 802 4 of 21

bench shown
testing. in Figure
A testing 1 isshown
bench established; battery
in Figure 1 is holders,
established;cables or connectors
battery are standard
holders, cables components
or connectors are
testing. A testing bench shown in Figure 1 is established; battery holders, cables or connectors are
of the battery
standard management
components system.
of the batteryIt is composed of
management a Battery
system. It isTest Systemof
composed 4000 (BTS4000,
a Battery TestNEWARE,
System
standard components of the battery management system. It is composed of a Battery Test System
4000 (BTS4000,
Shenzhen, NEWARE,
China) battery testerShenzhen,
with eight China) battery tester
independent channels,withaneight independent
intermediate channels,
machine, and a
4000 (BTS4000, NEWARE, Shenzhen, China) battery tester with eight independent channels,
an intermediate
64-bit host computer machine,
onto whichand aa 64-bit
NEWARE host BTS
computer
Softwareontov7.5.6
which anda NEWARE
a MATLAB BTS Software
R2012a v7.5.6
are installed.
an intermediate machine, and a 64-bit host computer onto which a NEWARE BTS Software v7.5.6
Theand a MATLAB
battery tester isR2012a
used toare installed.
load The batterycurrent
the programmed tester is used toon load the programmed
cells within acurrent
and a MATLAB R2012a are installed. The battery tester isprofiles used to load battery
the programmed range of
current
profiles
voltage on
of on
0~10 battery
V and cells within
a current a range
−6~6 of voltage of 0~10 V and a current of −6~6 A. The intermediate
profiles battery cells within aofrange ofA. The intermediate
voltage of 0~10 V andmachine
a current is ofmainly
−6~6 A. responsible
The intermediatefor the
machine
network is mainly responsible for control
the network connection, receiving the control controlling
command from the
machine is mainly responsible for the network connection, receiving the control command from lower
connection, receiving the command from the host computer, the the
host computer,
battery cycler andcontrolling
uploadingthe thelower battery
real-time testcycler
data.andTheuploading
host computerthe real-time
is employedtest data. The host
to control and
host computer, controlling the lower battery cycler and uploading the real-time test data. The host
computer
monitor is employed
the cycler to control
via Ethernet and as
cables, monitor
well as the
tocycler via voltage
store the Ethernet cables, as well as toacquired
store the
computer is employed to control and monitor the cycler via Ethernetand current
cables, as wellvalues
as to store theby
voltage and current values acquired by sensors. Batteries, brackets, cables, or connectors, as shown
sensors.
voltageBatteries,
and current brackets,
valuescables,
acquired or by
connectors, as shownbrackets,
sensors. Batteries, in Figure 1, areorstandard
cables, connectors,components
as shown of
in Figure 1, are standard components of the battery management system. The tested battery cells are
theinbattery
Figure management
1, are standardsystem.
componentsThe tested
of thebattery
battery cells are Panasonic
management system. NCR18650BE
The tested batterycylindrical cells
cells are
Panasonic NCR18650BE cylindrical cells (LiCoO2 oxide) whose specifications mainly include the
Panasonic
(LiCoO 2 oxide) NCR18650BE
whose cylindrical
specifications cells
mainly (LiCoO
include
2 oxide)
the whose
nominal specifications
capacity 3200 mainly
mAh, include
the the
operating
nominal capacity 3200 mAh, the operating voltage 3.7 V, the charging limit voltage 4.2 V, the
nominal
voltage 3.7 capacity 3200 mAh,
V, the charging limitthe operating
voltage 4.2 V,voltage 3.7 V, the
the discharge charging
limit voltagelimit 2.7 Vvoltage
and the4.2maximum
V, the
discharge limit voltage 2.7 V and the maximum continuous discharging current 10 A. During the
discharge
continuous limit voltage 2.7 V and the maximum continuous discharging
discharging current 10 A. During the test, the battery current and the voltage values current 10 A. During theare
test, the battery current and the voltage values are recorded every second, while they are recorded
test, theevery
recorded battery current
second, and the voltage
while values every
are recorded every second, while theyfor
are recorded
every 10 ms during the PPC they arefor
test. As recorded
the vibrating 10 ms3during
table, db/oct the
wasPPCappliedtest.inAs the vibrating
this experiment;
every
table, 10 ms during the PPC test. As for the vibrating table, 3 db/oct was applied in this experiment;
the 3Switch-2
db/oct was applied
(SW-2) in thiscontroller
vibration experiment; andthetheSwitch-2
MMF (Metra(SW-2) Mess-
vibration undcontroller and the MMF
Frequenztechnik in
the Switch-2
(Metra Mess- und (SW-2) vibration controller
Frequenztechnik in Radebeul ande.K.)
the vibration
MMF (Metra Mess- und
accelerometer were Frequenztechnik
also used. Basedin
Radebeul e.K.) vibration accelerometer were also used. Based on the test bench, the detailed teston
Radebeul
thesteps
test bench, e.K.) vibration
theelectrical accelerometer
detailed performance
test steps about were also used.
the electrical Based
performanceon the test bench, the detailed test
about the of cells are presented in Figureof 2. cells are presented in Figure 2.
steps about the electrical performance of cells are presented in Figure 2.

Figure 1. Battery Test System 4000 (BTS4000) power battery testing bench.
Figure1.1.Battery
Figure BatteryTest
TestSystem
System 4000
4000 (BTS4000) powerbattery
(BTS4000) power batterytesting
testingbench.
bench.

Figure 2. Schematic of electrical performance test process.

Figure2.2.Schematic
Figure Schematic of
of electrical
electrical performance
performancetest
testprocess.
process.
Appl. Sci. 2017, 7, 802 5 of 21

In pre-test, from Figure 2, constant current and constant voltage (CC-CV) charge and CC-discharge
are performed first in order to allow new cells to achieve a stable state of the performance; the
test data will therefore be more accurate. Between the two steps, cells were allowed to stabilize
during a rest period of 4 h [26]. It was identified within Ref. [21] that cells conditioned to 75% state
of charge (SOC) displayed a greater tendency for electrical degradation; therefore, each cell was
allocated a sample reference number, pre-conditioned to 75% SOC. In order to reduce the measurement
uncertainty of capacity and DC resistance, the cells’ characteristic values are used by the average of
three measurement results.
Post testing, if no significant mechanical damage or degradation (such as electrolyte leakage
or external fatigue cracking) was observed on any of test samples after the vibration test, then the
DC resistance, capacity, charge and discharge curve were measured at the environment temperature
ranging from 25 ◦ C ± 3 ◦ C.

2.2.2. Measuring the 1C Discharge Capacity

From the 100% SOC to the 0% SOC, the 1C constant current discharge to the cut-off voltage of
2.7 V to measure the cell capacity is used while recording the cell discharge curve.

2.2.3. Measurement Uncertainty of DC Resistance

Some factors that can affect testing reliability are cell temperature, calendar aging, etc. Hence,
the definition of internal resistance and an accurate measurement of internal resistance is very
important [27]. According to Ref. [28], the hybrid pulse power characterization (HPPC) test is used to
measure the DC resistance, the calculation formula of which is as follows:

U1 − U2
R= , (2)
I
where U1 represents the no-load voltage and U2 represents the voltage of 100 ms measurement time
after the voltage drop, and I represents the discharge [28].
An accurate measurement of internal resistance is critical, thus Equation (2) was used in this
paper for bias estimation. The first item is the method error, the second item and the third item are the
instrument error, and the fourth item is the environment error. The SOC, state of health (SOH), current,
temperature and time of measurement can affect the accuracy and consistency of the measurements [29].
In this paper, all internal resistances are measured in the 75% SOC, the discharge current is 1C, the
ambient temperature is 25 ◦ C ± 3 ◦ C, and the time of measurement is 100 ms; therefore, the influence
of these factors on internal resistance will be greatly reduced. As for SOH, the cycle life of the test
battery is greater than 1000; therefore, the effects of a few cycles on the internal resistance measurement
are not very large. Therefore, in this paper, the internal resistance measurement is very accurate.
The measurement uncertainty of resistance can be calculated by the following equation [28]:
     
 U1 − U2  1 1
∆R = 2

2
 |∆I | +  |∆U1 | +  |∆U2 | + | Tc ||∆T |,
  (3)
I  I I

where the ∆I, ∆U1 and ∆U2 measurement uncertainty are caused by measuring instruments; the
accuracy of the battery test system used in this test is 0.01% full scale (FS). The ∆T measurement
uncertainty are caused by the ambient temperature; the effect of this part can be neglected in this
experiment as the ambient temperature remains unchanged. Consequently, the results of ∆I, ∆U1 and
∆U2 calculations are 0.6 mA, 1 mV and 1 mV, respectively, and the value of ∆R is about 0.625 mΩ.
Eventually, the DC resistance can be written in (R ± ∆R) format.
 Appl. Sci. 2017, 7, 802 6 of 21
Appl. Sci. 2017, 7, 802 6 of 21

2.3. Design of Vibration Test

2.3. Design
Both of Vibration
United Test Recommendations on the Transport of Dangerous Goods (UN) 38.3 and
Nations
SAEBoth
J2380United
are common
Nations standards for the cell vibration
Recommendations and the of
on the Transport shock test. TheGoods
Dangerous UN 38.3 standard
(UN) was
38.3 and
chosen within Ref. [17]; while regulation UN 38.3 is a legal requirement, the
SAE J2380 are common standards for the cell vibration and the shock test. The UN 38.3 standard was standard is primarily
focused
chosen on emulating
within Ref. [17];the worst
while case shock
regulation UNload
38.3and
is a vibration that may be
legal requirement, theexperienced
standard is by the cell.
primarily
Load profiles are unrepresentative of those that would be experienced by the
focused on emulating the worst case shock load and vibration that may be experienced by the cell. battery during road
vehicle
Load operation
profiles [19]. However, it
are unrepresentative of isthose
noteworthy
that wouldthatbe SAE J2380 vibration
experienced by the profiles definedroad
battery during have
been synthesized from actual rough road measurement data and sequenced to emulate
vehicle operation [19]. However, it is noteworthy that SAE J2380 vibration profiles defined have been 100,000 miles
of vehicle usage.
synthesized In addition,
from actual rough the
roadUN 38.3, ECE Regulation
measurement 100, SAE J2380
data and sequenced and BS100,000
to emulate 62660 standards
miles of
were analyzed
vehicle usage. In within
addition,Ref.
the[19];
UN the
38.3,conclusion of the 100,
ECE Regulation SAESAE J2380 standard
J2380 and BScould
62660 better
standardsreflect
werethe
analyzed within Ref. [19]; the conclusion of the SAE J2380 standard could better reflect the actualis
actual vibration of lithium-ion cells which was obtained, and therefore the SAE J2380 standard
used in this
vibration paper. Thecells
of lithium-ion test which
steps ofwastheobtained,
vertical direction (i.e., the
and therefore the Z-axis) are standard
SAE J2380 summarized in Table
is used in this1,
and loading
paper. The testcurves
steps are presented
of the vertical in Figure 3(i.e., the Z-axis) are summarized in Table 1, and loading
direction
curves are presented in Figure 3
Table 1. Test schedule.
Table 1. Test schedule.
Test Conditions
Test Conditions Normal Test Normal Test Normal Test
State of Charge
Vibration Spectrum
Test Conditions Test (StableOC)
Conditions State Accel. (g rms) Time (h) Cumul. Time (h)
(%) of Normal Test Normal Test Normal Test
Vibration Spectrum Charge (StableOC) (%) Accel. (g rms) Time (h) Cumul. Time (h)
Vertical 1 spectrum 75 1.9 0.15 0.15
Vertical 1 spectrum 75 1.9 0.15 0.15
Vertical 1 spectrum 75 0.95 3.5 3.65
Vertical 1 spectrum 75 0.95 3.5 3.65
Vertical
Vertical2 2spectrum
spectrum 75
75 1.9 1.9 0.15 0.15 3.8 3.8
Vertical
Vertical2 2spectrum
spectrum 75
75 0.950.95 3.5 3.5 7.3 7.3
Vertical3 3spectrum
Vertical spectrum 75
75 1.9 1.9 0.15 0.15 7.45 7.45
Vertical3 3spectrum
Vertical spectrum 75
75 0.950.95 3.5 3.5 10.9510.95

0.12
Vert 1
Vert 2
Power Spectral Density(g2/Hz)

0.1
Vert 3
0.08

0.06

0.04

0.02

0
0 50 100 150 200
Frequency(Hz)

Figure 3. Vibration spectra for random vibration test referring to Society of Automotive Engineers
Figure 3. Vibration spectra for random vibration test referring to Society of Automotive Engineers
(SAE) J2380 standard.
(SAE) J2380 standard.

In this paper, a short-term vibration effect on the electrical properties of batteries was
In this paper,
researched. On the a short-term vibration
premise that effect on the
no resonance electrical
occurs duringproperties of batteries
the vibration wasthe
process, researched.
test was
On the premise that no resonance occurs during the vibration process, the
carried out based on the SAE J2380 standard, but some steps in the actual process are different test was carried out based
on the SAE J2380 standard, but some steps in the actual process are different
compared to the standard. For example, the standard requires the SOC at 100% and 20% to be compared to the standard.
For example,
complete, thea standard
but 75% SOCrequires
was used the SOC at 100%
according to theand 20% to [23].
literature be complete, but athe
In addition, 75% SOC was
standard used
requires
according to the literature [23]. In addition, the standard requires both X-axis
both X-axis and Y-axis direction to be used to conduct a certain vibration test, but it was noted and Y-axis direction to
be used to conduct a certain vibration test, but it was noted within [9,17,21] that
within [9,17,21] that the Z-axis vibration displayed a greater tendency for electrical degradation. the Z-axis vibration
displayed
The Z-axis a greater
profiletendency
was onlyfor electrical
loaded degradation.
in this The Z-axis
paper, in order profile was
to eliminate only loaded
the influence in this
of the paper,
non-main
invibration
order to eliminate the influence of the non-main vibration direction on the
direction on the electrical performance. However, it should be noted that, even if the electrical performance.
However, it should
battery will be noted
be affected in athat,
singleeven if the battery
vibration directionwillbut
be three
affected in a single
directions arevibration
subjecteddirection but
to vibration,
three directions
this does are subjected
not necessarily to an
have vibration,
impact;this does not
therefore, necessarily
it can be said have an contents
that the impact; therefore, it can be
of Refs. [21,23] are
said that the contents of Refs. [21,23] are studied in detail. In addition, the
studied in detail. In addition, the test only considers the effect of short-term vibration ontest only considers the effect
the
ofelectrical
short-term vibration onofthethe
performance electrical performance
cell, and the test ofoflong-term
the cell, and the test of
vibration is long-term
a key point vibration is a
of further
Appl. Sci. 2017, 7, 802 7 of 21

Appl. Sci. 2017, 7, 802 7 of 21

key point of further research. The definition of directions is presented in Figure 4. The Z-direction
research.
Appl. Sci. 2017,The definition of directions is presented in Figure 4. The Z-direction defined by the battery
7, 802
defined by the battery is in accordance with the Z-direction defined by the electric vehicle. 7 of 21
is in accordance with the Z-direction defined by the electric vehicle.
research. The definition of directions is presented in Figure 4. The Z-direction defined by the battery
is in accordance with the Z-direction defined by the electric vehicle.

Figure 4. Axis convention of cells.

Figure 4. Axis convention of cells.
3. Experimental Results
3. Experimental Results Figure 4. Axis convention of cells.
3.1. Effect of Vibration on 1C Discharge Capacity
3.1. Effect of VibrationResults
3. Experimental on 1C Discharge Capacity
3.1.1. Probability Density Function of Capacity Samples
3.1.1. 3.1.
Probability
Effect Density
of Vibration
In this paper, Function
32on of Capacity
1C Discharge
lithium-ion Capacity
cells Samplesof which the 1C capacity before vibration are
are investigated,
In within the range
thisProbability
paper, of 2930 mAh cells
32 lithium-ion to 3065
aremAh and the 1C capacity
investigated, of whichafter
thevibration are within
1C capacity the vibration
before range are
3.1.1.
of 2854 mAh to Density
3016 mAh.Function oftoCapacity
In order obtain theSamples
distribution of 1C capacity of samples before and
within the range of 2930 mAh to 3065 mAh and the 1C capacity after vibration are within the range
after
Invibration,
this paper,samples are grouped
32 lithium-ion inare
theinvestigated,
first place andofthe number
theof
1Cgroups is determined by the are
of 2854equation
mAh toas3016 mAh.
follows: In ordercells
to obtain which
the distribution of 1C capacity
capacity before vibration
of samples before and
within the range of 2930 mAh to 3065 mAh and the 1C capacity after vibration are within the range
after vibration, samples are grouped in the first place and the number of groups is determined by the
of 2854 mAh to 3016 mAh. In order to obtain m ≥the distribution
1+3.3 lgn , of 1C capacity of samples before (4) and
equation as follows:
after vibration, samples are grouped in the first place and the number of groups is determined by the
where n represents the number of testedmbattery ≥ 1 +samples
3.3lgn, (32). Therefore, the capacity of samples (4)
equation as follows:
should be divided into six groups.
where n represents
The medianthevalue
number of tested
of each is min≥ accordance
group battery samples
1+3.3 (32). the
lgn , with Therefore,
abscissa, the
and capacity of samples
the observed (4)
should probability
be divideddensity of each
into six group as the ordinate is calculated as follows:
groups.
where n represents the number of tested battery samples (32). Therefore, the capacity of samples
The median value of each group is in accordanceni with the abscissa, and the observed probability
should be divided into six groups. fn ( c ) = , (5)
density ofThe
eachmedian
group as the of
value ordinate is calculated
each group as
is in accordance follows:
nΔc i with the abscissa, and the observed
probability density of the
where ni represents eachnumber
group as
of the ordinate
samples is calculated
in each ngroup, n as follows: the number of the total
represents
i
samples and ∆ represents group distance. f n (c) = , (5)
n∆c
n i
fn ( c ) = iof ,the capacity before and after vibration is(5)
Based on Equation (5), the frequency histogram
nΔci
where nobtained; then, according to the sample distribution, the normal distribution probability density
i represents the number of samples in each group, n represents the number of the total samples
and ∆c curve
where is also obtained,
ni represents
represents groupthe as shown in Figures 5 and 6 respectively.
number of samples in each group, n represents the number of the total
distance.
i
samples
Based onand ∆ represents
Equation group
0.03 the
(5), distance.
frequency histogram of the capacity before and after vibration is
Based on Equation (5), the Frequency histogram
frequency histogram of the capacity before and after vibration is
obtained; then, according 0.025
to the sample distribution,
Normal distribution the normal distribution probability density curve
density curve
obtained; then, according to the sample distribution, the normal distribution probability density
is also obtained, as shown0.02in Figures 5 and 6 respectively.
curve is also obtained, as shown in Figures 5 and 6 respectively.
fn (c)

0.015
0.03
Frequency histogram
0.01
0.025 Normal distribution density curve

0.005
0.02

0
fn (c)

0.015 2850 2900 2950 3000 3050

Capacity(mAh)
0.01
Figure 5. The frequency histogram and theoretical normal distribution density function of the
capacity samples0.005
before vibration.

0
2850 2900 2950 3000 3050
Capacity(mAh)

Figure 5. The frequency histogram and theoretical normal distribution density function of the
Figure 5. The frequency histogram and theoretical normal distribution density function of the capacity
capacity samples before vibration.
samples before vibration.
Appl. Sci. 2017, 7, 802 8 of 21
Appl. Sci. 2017, 7, 802 8 of 21

0.025
Frequency histogram
Normal distribution density curve
0.02

0.015

fn (c) 0.01

0.005

0
2850 2900 2950 3000 3050
Capacity(mAh)

Figure 6. The
Figure 6. Thefrequency
frequency histogram
histogram andand theoretical
theoretical normal
normal distribution
distribution density
density function
function of the
of the capacity
capacity
samples samples after vibration.
after vibration.

From
From the above figures,
the above figures,ititcan
canbe
beseen
seenthat
thatthe
thefrequency
frequency histogram
histogram is well
is well in accordance
in accordance with
with the
the corresponding standard normal distribution. In more detail, the frequency histogram
corresponding standard normal distribution. In more detail, the frequency histogram before vibration before
vibration is in accordance
is in accordance with distribution
with the normal the normal N distribution N (2988.9,
(2988.9, 24.1794 2 ), and 24.1794 2), and the frequency
the frequency histogram after
histogram
vibration isafter vibration iswith
in accordance in accordance
the normalwith the normal
distribution distribution
N (2920.1, 29.3257N2(2920.1,
). 29.32572).

3.1.2.
3.1.2. Cumulative
Cumulative Distribution
Distribution Function
Function of
of Capacity
Capacity Samples
Samples
The
The theoretical
theoretical cumulative
cumulative distribution
distribution function
function of
of the
the capacity
capacity can be defined
can be defined as
as follows:
follows:

F (c() )= p{{C < }c}, ,

F c =p C<c (6)
(6)
where C represents the capacity of the samples and c represents the capacity of any cells.
where C represents the capacity of the samples and c represents the capacity of any cells.
Given the capacity sample , … in ascending order, the sample cumulative distribution
Given the capacity sample c1 , c2 . . . cn in ascending order, the sample cumulative distribution
function of the 1C capacity can be determined as the following:
function of the 1C capacity can be determined as the following:
  0,c < c1
 i 0, c < c1
 
F ( ( )
c ) = i  Δci f n ( c ) ,ci ≤ c < ci +1 , i = 1,2,...n-1 .

(7)
F (c) = ∑ (i =∆c
1 i f n ( c )), ci ≤ c < ci +1 , i = 1, 2, ...n − 1 . (7)
 i=1

1 ,c ≥ c
  1, c ≥ ncn
According to the Glivenko–Cantelli theorem, when the sample size is large, the maximum
According to the Glivenko–Cantelli theorem, when the sample size is large, the maximum
difference between the empirical cumulative distribution function ( ) ( ) and the theoretical
difference between the empirical cumulative distribution function F(n) (c) and the theoretical
cumulative distribution function ( ) would be small; in other words, ( ) can be displaced by
cumulative distribution function F (c) would be small; in other words, F (c) can be displaced by
estimating ( ) ( ).
estimating F(n) (c).
Based on the method described above, the empirical cumulative distribution function of the of
Based on the method described above, the empirical cumulative distribution function of the
samples for 1C capacity before and after vibration is obtained, and the cumulative function of
of samples for 1C capacity before and after vibration is obtained, and the cumulative function of
theoretical normal distribution before and after vibration is also obtained, as shown in Figures 7
theoretical normal distribution before and after vibration is also obtained, as shown in Figures 7
and 8, respectively.
and 8, respectively.
It can be seen from the above figures that the empirical cumulative distribution function is well
It can be seen from the above figures that the empirical cumulative distribution function is well in
in accordance with the cumulative function of the theoretical normal distribution, no matter
accordance with the cumulative function of the theoretical normal distribution, no matter whether it is
whether it is based on samples before vibration or based on the samples after vibration.
based on samples before vibration or based on the samples after vibration.
 Appl. Sci. 2017, 7, 802 9 of 21
Appl.
Appl. Sci.2017,
Sci. 2017,7,7, 802
802 9 of
9 of 2121

1
1 Sample distribution function
Sample distribution
Theoretical function
normal distribution
0.8 Theoretical normal distribution
0.8

0.6
(c)
FnF(c) 0.6
n

0.4
0.4

0.2
0.2

0
2800
0 2850 2900 2950 3000 3050
2800 2850 2900 Capacity(mAh)
2950 3000 3050
Capacity(mAh)

Figure7. 7.
Figure TheThe distribution
distribution function
function and theoretical
and theoretical normalnormal distribution
distribution functionfunction of the samples
of the capacity capacity
Figure 7. The distribution function and theoretical normal distribution function of the capacity
samples
before before
vibration. vibration.
samples before vibration.

1
1 Sample distribution function
Sample distribution
Theoretical function
normal distribution
0.8 Theoretical normal distribution
0.8

0.6
0.6
(c)
FnF(c)
n

0.4
0.4

0.2
0.2

0
2800
0 2850 2900 2950 3000 3050
2800 2850 2900 Capacity(mAh)
2950 3000 3050
Capacity(mAh)
Figure8. 8.
Figure TheThe distribution
distribution function
function and theoretical
and theoretical normalnormal distribution
distribution functionfunction of thesamples
of the capacity capacity
Figure 8. The distribution function and theoretical normal distribution function of the capacity
samples
after after vibration.
vibration.
samples after vibration.

3.1.3.Kolmogorov–Smirnov
3.1.3. Kolmogorov–Smirnov (K–S)Test Testforfor1C
1CCapacity
CapacityDistribution
Distribution
3.1.3. Kolmogorov–Smirnov(K–S) (K–S) Test for 1C Capacity Distribution
InIn thecase
Inthe
case wherethe
the casewhere
the populationdistribution
where thepopulation
distribution functionisiscompletely
population distributionfunction
completely unknownororonly
function is completelyunknown
only its form
unknown or onlyitsitsform form
known
known but its parameters unknown, the hypothesis test is generally performed in order
inferto infer
knownbut itsits
but parameters
parameters unknown,
unknown, the the
hypothesis
hypothesis test is generally
test performed
is generally performed in order in to
order tosome
infer
ofsome
some
of the properties
the properties
of the properties
of the population.
of the population.
of the population.
In this
In this study,
In thisthestudy,
K–S test
study,
theis K–S
the K–S
test is
adoptedtest to
adopted
is test whether
adopted
to test
the whether
to test capacity the
whether of
the
capacity
the samples of was
the samples
in was inwith
accordance accordance
the normalwith the normal Based
distribution. distribution.
on the Based
cumulative on thedistribution
cumulative
capacity of the samples was in accordance with the normal distribution. Based on the cumulative
distribution
Fn (c) function ( ) and the specified distribution function G(c), the tested as statistic
follows:is
function
distribution and
function ( ) and
the specified distribution function
the specified G(c), the tested
distribution statistic
function is constructed
G(c), the tested statistic is
constructed as follows:
constructed as follows:
((
Dn = max(| Fn (c) − G (c)|) ≤ Dn,α ,
Dn = max Fn ( c ) - G ( c ) ≤ Dn,α ,))
Dn = max Fn ( c ) - G ( c ) ≤ Dn,α ,
(8)
(8)
(8)
The null hypothesis is expressed as follows:
The null hypothesis is expressed as follows:
The null hypothesis is expressed as follows:

H00 :Fnn (( c)) = G(( c)) ,

H0 H: F:nF(c)c == G G(cc), , (9)
(9)
(9)
where D represents the critical value, n represents the number
where n represents the critical value, n represents the number of the total samples, and of the total samples, and α represents
where represents the critical value, n represents the number of the total samples, and
the significant
represents thelevel. For a given
significant level. value
For a of n and
given value α, byoflooking
n and up , bythelooking
quantile uptable of Kolmogorov
the quantile table of
represents the significant level. For a given value of n and , by looking up the quantile table of
distribution the critical value D
Kolmogorov distribution the critical value
Kolmogorov
than Dn,α , the null distribution
hypothesis
n,α can
theiscritical
accepted,
be obtained
value
otherwise
to satisfy P { D ≤ D }
, can be obtained to satisfy P Dn ≤ Dn,α
, can it isbe
rejected.
n n,α = 1 −
{{ α; then,
obtained to satisfy P Dn ≤ Dn,α = 1 − ;
}}
if D = is
n 1 less
− ;
then, if the capacity
is less than , , theofnull hypothesis is accepted, byotherwise it is rejected.
For
then, if is less than , , the null
distribution samples before vibration,
hypothesis is accepted, taking advantage
otherwise of MATLAB R2012a,
it is rejected.
For
Dn is obtained the capacity
as 0.1711,distribution
and for the of
givensamples before
significant vibration,
level = 0.05,byD taking advantage of 0.2379
MATLAB
For the capacity distribution of samples before vibration, by n,α α taking advantage as
is also obtained of MATLAB by
R2012a, is obtained as 0.1711, and for the given significant level = 0.05, , is also
looking
R2012a,at the table. Obviously,
is obtained it can beand
as 0.1711, seenfor
that the Dngiven than Dn,α , solevel
is less significant =hypothesis
the null 0.05, cannot
is also
obtained as 0.2379 by looking at the table. Obviously, it can be seen that is less than , , , so the
obtained as 0.2379 by looking at the table. Obviously, it can be seen that is less than , so the
,
 Appl. Sci. 2017, 7, 802 10 of 21
Appl. Sci. 2017, 7, 802 10 of 21
null hypothesis cannot be rejected. In other words, it is believed that the capacity distribution of
samples before vibration obeys the normal distribution N (2988.9, 24.17942).
be rejected. In other words, it is believed that the capacity distribution of samples before vibration
Similarly, for the capacity distribution of the samples after vibration, is obtained as 0.2065,
obeys the normal distribution N (2988.9, 24.17942 ).
and for the given significant level = 0.05 , , is obtained as 0.2379. Obviously, it can be also
Similarly, for the capacity distribution of the samples after vibration, Dn is obtained as 0.2065, and
seen that is less than , , so the null hypothesis cannot be rejected; in other words, it is also
for the given significant level α = 0.05 Dn,α , is obtained as 0.2379. Obviously, it can be also seen that Dn
believed that the capacity distribution of the samples after vibration obeys the normal distribution
is less than Dn,α , so2 the null hypothesis cannot be rejected; in other words, it is also believed that the
N (2920.1, 29.3257 ).
capacity distribution of the samples after vibration obeys the normal distribution N (2920.1, 29.32572 ).
3.1.4.Comparison
3.1.4. Comparisonfor forCapacity
CapacityDistribution
DistributionofofSamples
Samplesbefore
beforeand
andafter
afterVibration
Vibration
Accordingtoto
According thethe analysis
analysis results
results above,
above, the probability
the probability densitydensity curves
curves of of the for
the capacity capacity for
samples
samples
before andbefore and after are
after vibration vibration are superimposed
superimposed in Figure 9.
in Figure 9.

0.02
Pre-Test
Post-Test
0.015
f(c)

0.01

0.005

0
2800 2850 2900 2950 3000 3050 3100
Capacity(mAh)

Figure9.9.The
Figure The distribution
distribution function
function and theoretical
and theoretical normal
normal distribution
distribution functionfunction of the samples
of the capacity capacity
samples after
after vibration. vibration.

From Figure 9, it can be seen that the probability density curve after vibration exhibits a shift
From Figure 9, it can be seen that the probability density curve after vibration exhibits a shift to
to the left and is flat as well as wide compared to that before vibration. This shows that the
the left and is flat as well as wide compared to that before vibration. This shows that the capacity mean
capacity mean of samples exhibited a reduction by 68.8 mAh and the standard variation increased
of samples exhibited a reduction by 68.8 mAh and the standard variation increased by 5.1463 mAh
by 5.1463 mAh after vibration.
after vibration.
3.2.Effect
3.2. EffectofofVibration
Vibrationon
onDC
DCResistance
Resistance
TheDC
The DCresistance
resistancedistribution
distribution before
before and
and after
after the
thevibration
vibration test
test are
are presented
presented in
in Tables
Tables22
and 3, respectively.
and 3, respectively.

Table2.2.DC
Table DCresistance
resistancedistribution
distributionbefore
beforevibration
vibrationtest
test(mΩ).
(mΩ).

Test Condition DC Internal Resistance

Test Condition DC Internal Resistance
38.03 39.47 39.07 39.67 37.40 38.47 38.00
38.03 39.4736.6039.07
36.80 36.37 39.67 37.40 37.83
38.23 37.40 38.47 36.8038.00
36.80 36.60 36.37 38.23 37.40 37.83 36.80
Pre-Test 37.40 37.12 36.73 36.92 36.16 36.64 36.79
Pre-Test 37.40 37.12 36.73 36.92 36.16 36.64 36.79
36.23 36.23
34.9334.9337.56
37.56 37.27
37.27 36.66
36.66 36.67
36.67 37.6137.61
36.10 36.10
36.5236.5235.69
35.69

Table 3. DC resistance distribution after vibration test (mΩ).

Table 3. DC resistance distribution after vibration test (mΩ).

Test Condition DC Internal Resistance

Test Condition DC Internal Resistance
37.63 38.87 38.67 38.63 37.40 38.43 38.03
37.63 38.87
35.57 37.8338.67
37.20 38.63 37.40
38.43 37.83 38.43
38.87 37.2038.03
35.57 37.83 37.20 38.43 37.83 38.87 37.20
Post-Test38.03
Post-Test
38.03
38.2338.2337.60
37.60 38.83
38.83 36.60
36.60 37.43
37.43 37.6337.63
36.57 36.57
36.1736.1739.20
39.20 42.44
42.44 38.72
38.72 38.52
38.52 40.0940.09
38.43 38.5038.5037.13
38.43 37.13
Appl. Sci.
Appl. Sci. 2017,
2017, 7,
7, 802
802 11of
11 of21
21

According to existing studies, the effect of vibration on the electrical performance of the battery
According
is weak. Therefore, to existing studies, the
the conclusion effect
that of vibration
vibration has noon the electrical
effect performance
on the electrical of the cannot
properties battery be
is
weak. Therefore, the conclusion that vibration has no effect on the electrical
made, even if the electric performance does not change significantly before and after vibration. properties cannot be made,
even if the electric
The statistical performance
knowledge mustdoes not change
be used to studysignificantly before and
how the electrical after vibration.
properties change. The statistical
knowledge must be used to study how the electrical properties change.
As can be seen from Tables 2 and 3 and Figure 10, some cells exhibit an increase in internal
As canwhile
resistance be seen from
others Tablesa 2decrease
exhibit and 3 and Figure
after 10, sometest
the vibration cells exhibit
simply onantheincrease in internal
perspective of the
resistance while others exhibit a decrease after the vibration test simply on
single cell; therefore, there are no consistent conclusions to be reached on this problem as to whetherthe perspective of the
single cell; therefore, there are no consistent conclusions to be reached on this
the vibration and the mechanical loading lead to an increase or a decrease in the internal resistance. problem as to whether
the vibration
However, fromandanother
the mechanical loading
point of view, lead can
it also to an beincrease
seen that, or aondecrease
the whole,in the internal
there is an resistance.
increasing
However, from another point of view, it also can be seen that, on the whole,
trend in internal resistance, where 26 out of 32 cells which exhibit an increase in the internal there is an increasing
trend in internal
resistance occupy resistance, where
most of the 26 out
sample, of 32just
while cells6 which exhibit
out of 32 cells an increase
exhibit in the
a slight internal Therefore,
decrease. resistance
occupy
in ordermost of the sample,
to eliminate while justof
the influence 6 out
the ofmeasurement
32 cells exhibit a slight decrease.
uncertainty and other Therefore, in order
accidental to
factors
eliminate the influence of the measurement uncertainty and other accidental factors
which may lead to an illusion of a decreasing trend, the mathematical statistical analysis is adopted which may lead to
an illusion of a decreasing trend,
to excavate a more reasonable conclusion. the mathematical statistical analysis is adopted to excavate a more
reasonable conclusion.

44 Pre-Test
Post-Test
42
DC Resistance(mΩ)

40

38

36

34

32
0 5 10 15 20 25 30 35
Sample Number

Figure 10.
Figure 10. DC
DC resistance
resistance distribution
distribution before
before and
andvibration
vibrationtest.
test.

In order to identify whether there is an increasing trend that relates to the impact of vibration
In order to identify whether there is an increasing trend that relates to the impact of vibration on
on the cell resistance, this paper adopts the method of hypothesis testing to give reasonable
the cell resistance, this paper adopts the method of hypothesis testing to give reasonable assumptions.
assumptions. Based on the sample distribution of the internal resistance before and after the
Based on the sample distribution of the internal resistance before and after the vibration test, a decision
vibration test, a decision regarding the acceptance or rejection of the null hypothesis is made under a
regarding the acceptance or rejection of the null hypothesis is made under a certain significant level
certain significant level . In general, the null hypothesis is the hypothesis that the researcher wants
α. In general, the null hypothesis is the hypothesis that the researcher wants to collect the evidence
to collect the evidence to object to, which is expressed usually as H0. The alternative hypothesis is the
to object to, which is expressed usually as H0 . The alternative hypothesis is the hypothesis that the
hypothesis that the researcher wants to collect evidence to support this, which is expressed as H1.
researcher wants to collect evidence to support this, which is expressed as H1 . Therefore, in this
Therefore, in this paper, the null hypothesis is ≤ where μ represents the internal resistance
paper, the null hypothesis is µ ≤ µ0 where µ represents the internal resistance mean of samples
mean of samples before vibration testing and represents the internal resistance mean of samples
before vibration testing and µ0 represents the internal resistance mean of samples after vibration
after vibration testing. Consequently, the alternative hypothesis is > . If a decision to reject the
testing. Consequently, the alternative hypothesis is µ > µ0 . If a decision to reject the null hypothesis
null hypothesis is made under a certain significant level = 0.05, it would show that there is a
is made under a certain significant level α = 0.05, it would show that there is a probability of 95%
probability of 95% that the sample mean after the vibration test is greater than the sample mean
that the sample mean after the vibration test is greater than the sample mean before the vibration
before the vibration test. In other words, there are good reasons to believe it is vibration which leads
test. In other words, there are good reasons to believe it is vibration which leads to an increase in the
to an increase in the internal resistance. Therefore, hypothesis testing can be used to explain the
internal resistance. Therefore, hypothesis testing can be used to explain the effect of the vibration on
effect of the vibration on the internal resistance of the battery in a statistically significant way. Before
the internal resistance of the battery in a statistically significant way. Before hypothesis testing, it is
hypothesis testing, it is necessary to perform a parameter estimation of the sample.
necessary to perform a parameter estimation of the sample.
3.2.1. Parameters
3.2.1. Parameters Estimation
Estimation Based
Based on
on Samples
Samples before
beforeVibration
Vibration
The resistance
The resistance of
of the
the battery
battery before
before the
the vibration
vibration is
is regarded
regarded asas aa sample
sample ofof this
this batch
batch ofof batteries,
batteries,
the mean value and the standard deviation of which are estimated by the
the mean value and the standard deviation of which are estimated by the method of parametermethod of parameter
estimation. In
estimation. In the
the next
next section,
section, the
the internal
internal resistance
resistance of
of the
the samples
samples after
after vibration
vibration is
is tested
tested with
with the
the
standard by which the parameters are estimated based on the samples before
standard by which the parameters are estimated based on the samples before vibration. vibration.
Before the parameter estimation, it is necessary to test the population distribution of the
internal resistance of samples before vibration in the same way as described in Section 3.1 for the
Appl. Sci. 2017, 7, 802 12 of 21

Before the parameter estimation, it is necessary to test the population distribution of the internal
resistance of samples before vibration in the same way as described in Section 3.1 for the capacity
distribution test; that is to say, the frequency histogram is fitted at first, then the K–S hypothesis
testing is carried out, and finally the distribution of the internal resistance based on the samples before
vibration is obtained as normal distribution N (37.20, 1.0562 ).

Estimation of Confidence Interval for the Population Mean µ0

According to the theorem related to statistics, suppose the sample values obey the population
normal distribution N (µ, σ2 ), applying the following formula:

X−µ
√ ∼ t ( n − 1), (10)
S/ n

where X, S2 denotes the sample mean and sample variance, respectively, and t(n − 1) does not depend
on any unknown parameters.
X −µ
Using S/√n as the statistic, for a given confidence level 1 − α, the following formula applies:

X−µ
 
P − t α ( n − 1) < √ < t α (n − 1) = 1 − α. (11)
2 S/ n 2

Thus, the confidence interval of the population mean µ0 is obtained as

S
X ± √ t α ( n − 1). (12)
n 2

When the confidence level 1 − α is set to be 0.95, by taking advantage of MATLAB R2012a, the
confidence interval of the overall mean µ0 is calculated, resulting in (36.81, 37.58) mΩ.

Estimation of Confidence Interval for the Population Standard Variation σ0

According to the theorem related to statistics, suppose the sample values obey the population
normal distribution N (µ, σ2 ), applying the following formula:

( n − 1) S2
∼ χ2 ( n − 1), (13)
σ2

where X, S2 denotes the sample mean and the sample variance, respectively, and χ2 (n − 1) does not
depend on any unknown parameters.
(n−1) S2
Using σ2
as the statistic, for a given confidence level 1 − α, the following formula applies:

( n − 1) S2
 
P χ21− α (n − 1) < < χ2α (n − 1) = 1 − α. (14)
2 σ2 2

Thus, the confidence interval of the population standard variation σ0 is obtained as


√ √ 
n − 1S n − 1S
q , 2 . (15)
χ2α (n − 1) χ1− α2 (n − 1)
2

When the confidence level 1 − α is set to be 0.95, by taking advantage of MATLAB R2012a, the
confidence interval of the population standard variation σ0 is calculated, resulting in (0.844, 1.411) mΩ.
Appl. Sci. 2017, 7, 802 13 of 21

3.2.2. Hypothesis Testing for Samples after Vibration

Similarly, before the hypothesis testing, it is necessary to test the population distribution of the
internal resistance based on samples after vibration in the same way as described in Section 3.1 for the
capacity distribution test. Finally, the distribution of the internal resistance based on the samples after
vibration is obtained as normal distribution N (38.09, 1.2442 ).

Hypothesis Testing for the Internal Resistance Mean µ after Vibration

Confidence intervals of the population mean µ0 and the standard variation σ0 have been obtained
in the previous section. According to the theorem related to statistics, based on the normal distribution
2
N (µ0max , σ0max ) the Z-test is adopted to test whether the internal resistance mean µ after vibration
exhibits a significant increase compared to that before vibration. Therefore, the null hypothesis and
the alternative hypothesis are expressed as follows:
The null hypothesis H0 :
µ ≤ µ0max . (16)

The alternative hypothesis H1 :

µ > µ0max . (17)
X −µ
Then, the statistic Z = σ 0max √ is adopted to determine the rejection region. Thus, the rejection
0max / n
region can be expressed as follows:

X − µ0max
Z= √ ≥ Zα , (18)
σ0max / n

where n = 32, X = 38.09, µ0max = 37.58 and σ0max = 1.411, respectively, from the sections above.
Based on Equation (18), Z = 2.0023 is obtained and then, according to the standard normal
distribution table, the corresponding P is obtained as 0.0226. Given the significant level α = 0.05, there
is P < α. Therefore, there are good reasons to refuse the null hypothesis, and it is accepted that the
internal resistance mean µ after vibration increased at significant level α = 0.05.

Hypothesis Testing for the Resistance Standard Variation σ after Vibration

According to the theorem related to statistics, the chi-square test is adopted to test whether the
internal resistance standard variation σ after vibration exhibits a significant increase compared to
that before vibration. Therefore, the null hypothesis and the alternative hypothesis are expressed
as follows:
The null hypothesis H0 :
σ2 ≤ σ02 . (19)

The alternative hypothesis H1 :

σ2 > σ02 . (20)
(n−1) S2
Then, the statistic χ2 = σ02
is adopted to determine the rejection region. Thus, the rejection
region can be expressed as follows:

( n − 1) S2
χ2 = ≥ χ2α (n − 1), (21)
σ02

where n = 32, S = 1.244 and σ0 = 1.056, respectively, from the above sections.
Based on Equation (21), χ2 = 41.633 is obtained, and then, given the significant level α = 0.05, we
check the corresponding table and get the value χ2α (n − 1) = 44.98. Consequently, χ2 < χ2α (n − 1). In
other words, there are not good reasons to refuse the null hypothesis at the significant level α = 0.05.
Similarly, given the significant level of α = 0.1, we check the corresponding table and get the value
Appl. Sci. 2017, 7, 802 14 of 21
Appl. Sci. 2017, 7, 802 14 of 21

χ2α (n −the
refuse 1) =null
41.422. Consequently,
hypothesis,
2 > χ2 ( n − 1). Therefore, there are good reasons to refuse the null
and it χis accepted
α that the internal resistance standard variation σ after
vibration increased at significant level = 0.1.
hypothesis, and it is accepted that the internal resistance standard variation σ after vibration increased
at significant level α = 0.1.
4. Consistency Analysis
4. Consistency Analysis
4.1. Feature Extraction
4.1. Feature Extraction
4.1.1. Peaks Value of Incremental Capacity (IC) Curves Phase Transformation
4.1.1. Peaks Value of Incremental Capacity (IC) Curves Phase Transformation
The incremental capacity analysis (ICA) method has been applied to the study of the aging
The incremental capacity analysis (ICA) method has been applied to the study of the aging
mechanism of cells, and this method has the potential to be integrated in a battery management
mechanism of cells, and this method has the potential to be integrated in a battery management system
system (BMS) to estimate the cell SOH and observe the internal changes in cells in real time [30].
(BMS) to estimate the cell SOH and observe the internal changes in cells in real time [30]. Cell IC curves
Cell IC curves could be determined on the basis of the constant current charge curves.
could be determined on the basis of the constant current charge curves.
During the cell charging or discharging process, it can be easily determined that, with the insertion
During the cell charging or discharging process, it can be easily determined that, with the insertion
and extraction of lithium ions, cathode and anode materials will undergo phase transformation, and
and extraction of lithium ions, cathode and anode materials will undergo phase transformation, and
such a phase transformation corresponds to a voltage plateau. The IC curve is defined as follows.
such a phase transformation corresponds to a voltage plateau. The IC curve is defined as follows.
dQ d ( It ) I
P ( V ) = dQ = d(It) = I . (22)
P(V ) = dV = dV =dV / dt . (22)
dV dV dV/dt
In this paper, the IC curve has been smoothed, so the decrease of capacity may not very precise,
In this paper, the IC curve has been smoothed, so the decrease of capacity may not very precise,
but it can indicate the trend of the voltage value of the voltage plateau increasing and the capacity
but it can indicate the trend of the voltage value of the voltage plateau increasing and the capacity
decreasing. Figure 11 shows the IC curves of Cell 1 before and after the vibration test; there are three
decreasing. Figure 11 shows the IC curves of Cell 1 before and after the vibration test; there are
obvious voltage plateaus, but the cell IC curve after the vibration test has a tendency to move right as
three obvious voltage plateaus, but the cell IC curve after the vibration test has a tendency to move
a whole, and the peak value is slightly lower than before the vibration test. According to the
right as a whole, and the peak value is slightly lower than before the vibration test. According to
characteristics of the cell aging IC curve, the position of the voltage plateau shift to the right and the
the characteristics of the cell aging IC curve, the position of the voltage plateau shift to the right and
peak value drop shows that there is degradation of cell electrical performance. In this paper,
the peak value drop shows that there is degradation of cell electrical performance. In this paper, the
the voltage value of the voltage plateau is regarded as a characteristic quantity to evaluate the
voltage value of the voltage plateau is regarded as a characteristic quantity to evaluate the consistency
consistency of the cells; as a result, three characteristic quantities can be found, which are denoted as
of the cells; as a result, three characteristic quantities can be found, which are denoted as P1, P2 and
P1, P2 and P3, respectively.
P3, respectively.

8
P2
Pre-Test
P1 Post-Test
6
dQ/dV(Ah.V-1)

P3
4

0
3.5 3.6 3.7 3.8 3.9 4 4.1 4.2 4.3
Voltage(V)

Figure
Figure 11.
11. Incremental
Incremental capacity
capacity (IC)
(IC) curves before and after vibration.

4.1.2.
4.1.2. Ratio
Ratio of
of Potentio-Charge
Potentio-Charge Capacity
Capacity to
to Galvano-Charge Capacity
Galvano-Charge Capacity
Here
Here aa method
method based
based onon the
the ratio
ratio of
of the
the potentio-charge
potentio-charge capacity
capacity toto the
the galvano-charge
galvano-charge capacity
capacity
(RPG) was developed to determine the diffusion coefficient of intercalary species
(RPG) was developed to determine the diffusion coefficient of intercalary species for porous electrode for porous
electrode by the diffusion
by the spherical spherical model
diffusion
[31].model
In the[31]. In the
process of process
movement,of movement,
the electronstheand
electrons
ions inand ions in
the battery
the
willbattery willtobecertain
be subject subjectresistance,
to certainwhich
resistance, whichpolarization.
will cause will cause polarization.
With the polarization
With the polarization enhanced,
enhanced, the cell’s cycle
the cell’s cycle performance
performance will
will bebe worse.
worse. When
When the the cell
cell is
is
charged
charged at at aaconstant
constantcurrent,
current,thetheconcentration
concentrationpolarization
polarizationincreases
increases gradually
gradually with
withthethe
increase of
increase
the terminal
of the terminal voltage, and
voltage, andthethe
cell voltage
cell voltagereaches
reachesthetherated
ratedvalue
valuewhen
whenititisis not
not full.
full. During
During the the
constant voltage charging process, the concentration polarization is gradually eliminated,
constant voltage charging process, the concentration polarization is gradually eliminated, so that the so that the
cell capacity is full. It can be easily concluded that the more capacity charged of constant current,
Appl. Sci. 2017, 7, 802 15 of 21

Appl. Sci. 2017, 7, 802 15 of 21

cell capacity is full. It can be easily concluded that the more capacity charged of constant current, the
better cycle
the better performance
cycle performanceof the cell,cell,
of the or vice versa.
or vice Therefore,
versa. thethe
Therefore, RPG cancan
RPG be used as aascharacteristic
be used to
a characteristic
evaluate thethe
to evaluate consistency
consistencyof the cell,cell,
of the which is defined
which as follows:
is defined as follows:
C
qq== CCV
CV .
. (23)
(23)
C
CCC
CC

Figure
Figure 1212represents
represents thethe
voltage, current
voltage, and capacity
current curve curve
and capacity duringduring
the constant current constant
the constant current
voltage charging process of Cell 1 before and after the vibration test. From
constant voltage charging process of Cell 1 before and after the vibration test. FromFigure 10, the charging
Figuretime
10,
of the constant current after the vibration test is shorter than before the test, indicating
the charging time of the constant current after the vibration test is shorter than before the test, that there may
be a degradation
indicating of cell
that there may electrical performance.
be a degradation It can
of cell also be performance.
electrical seen that the time
It canofalso
the be
whole
seencharging
that the
process becomes
time of the wholelonger the process
charging after vibration
becomestest.
longer the after vibration test.

4.4
Pre-Test
Post-Test
4.2
Voltage(V)

3.8

3.6

3.4
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Time(s)
(a)
3.5
Pre-Test
3 Post-Test

2.5
Current(A)

1.5

0.5

0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Time(s)
(b)
3.5 Pre-Test
Post-Test
3

2.5
Capacity(Ah)

1.5

0.5

0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000
Time(s)
(c)
Figure 12. Constant current constant voltage charging process. (a) Voltage variation of the constant
Figure 12. Constant current constant voltage charging process. (a) Voltage variation of the constant
current and constant voltage (CC-CV) charging; (b) Current variation of the CC-CV charging;
current and constant voltage (CC-CV) charging; (b) Current variation of the CC-CV charging;
(c) Capacity variation
(c) Capacity variation of
of the
the CC-CV
CC-CV charging.
charging.

4.1.3. Discharge
4.1.3. Discharge Curve
Curve
The discharge
The discharge curve
curve hashas been
been used
used toto predict
predict capacity
capacity fading
fading of
of lithium-ion
lithium-ion cells
cells [32],
[32], which
which can
can
reflect the discharge cut-off point, discharge plateau and capacity of the cell; the discharge
reflect the discharge cut-off point, discharge plateau and capacity of the cell; the discharge curve can curve can
therefore be
therefore be used
used as
as aa feature
feature of
of the
the cell
cell consistency
consistency evaluation.
evaluation. In
In this
this paper,
paper, the
the area
area surrounded
surrounded
by the discharge curve is used to describe the similarity of the discharge curve, which is defined
as follows:
Appl. Sci. 2017, 7, 802 16 of 21

Appl. Sci. 2017, 7, 802 16 of 21

by the discharge curve is used to describe the similarity of the discharge curve, which is defined as
follows:

Z
AA== Udt
Udt.. (24)
(24)
Figure 13
Figure 13shows
shows discharge
discharge curves
curves before
before and the
and after after the vibration
vibration test1; the
test of Cell of Cell 1; the area
area surrounded
by the discharge curve after the vibration test is smaller than before the test, indicating that indicating
surrounded by the discharge curve after the vibration test is smaller than before the test, there may
that
be there may beofa cell
a degradation degradation
electricalof cell electrical performance.
performance.

4.5
Pre-Test
Post-Test
4
Voltage(V)

3.5

2.5
0 500 1000 1500 2000 2500 3000 3500 4000
Time(s)

Figure 13. Discharge

Figure 13. Discharge curves
curves before
before and
and after
after vibration
vibration test.
test.

In summary,
In summary, the the features
featuresofofthe
theconsistency
consistencyevaluation forfor
evaluation lithium-ion cells
lithium-ion are are
cells capacity (C), (C),
capacity DC
resistance
DC (R), (R),
resistance the the
position of the
position peaks
of the (P1, (P1,
peaks P2, P2,
P3) P3)
of the voltage
of the plateau,
voltage the the
plateau, ratioratio
(q) of
(q)the
of
potentio-charge
the potentio-charge capacity to the
capacity galvano-charge
to the galvano-charge capacity
capacityandandthethearea
area(A)
(A) surrounded
surrounded by the
by the
discharge curve.
discharge curve.

4.2.
4.2. Feature Calculation
The value
value of
of DC
DC resistances
resistances (R)
(R) and
and capacities
capacities (C)(C) are
are obtained
obtained by
by the
the HPPC
HPPC test test and
and constant
constant
current discharge test, respectively; the position of the peaks (P1, P2, P3) of of the
the voltage
voltage plateau
plateau are
are
obtained by drawing
drawing IC curves,
curves, the ratio
ratio (q)
(q) of
of the
the potentio-charge
potentio-charge capacity
capacity toto the
the galvano-charge
galvano-charge
capacity
capacity and
andthe
thearea (A)(A)
area surrounded by the
surrounded bydischarge curve are
the discharge obtained
curve via calculating
are obtained the formulae
via calculating the
which were
formulae described
which were in Section 4.1.
described The calculated
in Section 4.1. Theresults of each
calculated cell before
results of eachand after
cell vibration
before test
and after
are showntest
vibration in Tables 4 andin5.Tables 4 and 5.
are shown

Table 4. Consistency
Table 4. Consistency features
features of
of cells
cells sample
sample before
before vibration
vibration test.
test.

Sample
Sample NumberCapacity
Number Capacity (mAh) DC
(mAh) DC Resistance
Resistance (mΩ)
(mΩ) qq AA (Vs)
(Vs) P1
P1(V)
(V) P2 P2
(V)(V) P3 (V)
P3 (V)
1 1 2984.2
2984.2 38.03
38.03 0.2073
0.2073 12576.5
12576.5 3.615
3.615 3.648
3.648 4.15 4.15
2 2 2967.1
2967.1 39.47
39.47 0.2061
0.2061 12498.8
12498.8 3.624
3.624 3.644
3.644 4.1264.126
3 3 2987.9
2987.9 39.07
39.07 0.1975
0.1975 12582.7
12582.7 3.617
3.617 3.647
3.647 4.1294.129
4 4 2930.4
2930.4 39.67
39.67 0.2083
0.2083 12364.4
12364.4 3.64
3.64 3.66
3.66 4.1274.127
5 2964.1 37.40 0.2021 12488.5 3.622 3.654 4.126
6 5 2964.1
2965.7 37.40
38.47 0.2021
0.2050 12488.5
12491.7 3.622
3.622 3.654
3.654 4.1264.128
7 6 2965.7
2977.9 38.47
38.00 0.2050
0.2038 12491.7
12570.6 3.622
3.619 3.654
3.651 4.1284.165
8 7 2977.9
2978.7 38.00
36.80 0.2038
0.1970 12570.6
12559.1 3.619
3.622 3.651
3.656 4.1654.129
9 8 2978.2
2978.7 36.60
36.80 0.2070
0.1970 12567.5
12559.1 3.612
3.622 3.646
3.656 4.1294.137
11 3002.8 36.37 0.2016 12666.4 3.606 3.651 4.129
9 2978.2 36.60 0.2070 12567.5 3.612 3.646 4.137
12 2993.4 38.23 0.2024 12621.0 3.616 3.658 4.134
13 11 3002.8
2974.7 36.37
37.40 0.2016
0.2015 12666.4
12540.5 3.606
3.61 3.651
3.655 4.1294.123
14 12 2993.4
3023.8 38.23
37.83 0.2024
0.1953 12621.0
12739.2 3.616
3.614 3.658
3.655 4.1344.132
15 13 2982.7
2974.7 36.80
37.40 0.2018
0.2015 12582.8
12540.5 3.606
3.61 3.653
3.655 4.1234.134
16 14 3015.6
3023.8 37.40
37.83 0.1975
0.1953 12722.3
12739.2 3.61
3.614 3.647
3.655 4.1324.134
17 2986 37.12 0.2032 12584.3 3.6 3.647 4.147
18 15 2982.7
2964.3 36.80
36.73 0.2018
0.2063 12582.8
12503.1 3.606
3.599 3.653
3.636 4.1344.141
19 16 3015.6
3001 37.40
36.92 0.1975
0.1962 12722.3
12637.8 3.61
3.596 3.647
3.642 4.1344.123
20 17 2997.1
2986 36.16
37.12 0.1972
0.2032 12638.3
12584.3 3.6
3.6 3.646
3.647 4.1474.14
21 18 2986.2
2964.3 36.64
36.73 0.1971
0.2063 12588.7
12503.1 3.601
3.599 3.635
3.636 4.1414.142
22 2985.7 36.79 0.2003 12579.5 3.602 3.648 4.133
19 3001 36.92 0.1962 12637.8 3.596 3.642 4.123
23 2980.9 36.23 0.2000 12577.5 3.596 3.655 4.134
24 20 2997.1
2990.7 36.16
34.93 0.1972
0.1922 12638.3
12610.3 3.6
3.586 3.646
3.648 4.144.136
21 2986.2 36.64 0.1971 12588.7 3.601 3.635 4.142
22 2985.7 36.79 0.2003 12579.5 3.602 3.648 4.133
23 2980.9 36.23 0.2000 12577.5 3.596 3.655 4.134
24 2990.7 34.93 0.1922 12610.3 3.586 3.648 4.136
Appl. Sci. 2017, 7, 802 17 of 21

Table 4. Cont.

Sample Number Capacity (mAh) DC Resistance (mΩ) q A (Vs) P1 (V) P2 (V) P3 (V)
25 2979.1 37.56 0.2076 12564.5 3.607 3.644 4.139
26 2987.2 37.27 0.2033 12586.6 3.603 3.638 4.136
27 2988 36.66 0.1990 12587.7 3.602 3.648 4.132
28 3000.4 36.67 0.1950 12646.5 3.6 3.65 4.139
29 2997.4 37.61 0.1983 12626.9 3.607 3.651 4.141
30 2985.7 36.10 0.2007 12574.7 3.596 3.646 4.134
31 3045.4 36.52 0.1908 12836.2 3.6 3.636 4.134
32 3064.6 35.69 0.1852 12928.2 3.587 3.637 4.136

Table 5. Consistency features of cells sample after vibration test.

Sample Number Capacity (mAh) DC Resistance (mΩ) q A (Vs) P1 (V) P2 (V) P3 (V)
1 2925.9 37.63 0.2222 12331.6 3.646 3.677 4.151
2 2907.4 38.87 0.2200 12240.2 3.657 3.685 4.127
3 2927.6 38.67 0.2117 12320.8 3.651 3.678 4.133
4 2854.1 38.63 0.2228 12036.6 3.689 3.708 4.143
5 2913.9 37.40 0.2157 12264.1 3.655 3.684 4.141
6 2913.4 38.43 0.2208 12256.3 3.656 3.686 4.133
7 2920.5 38.03 0.2190 12319.9 3.651 3.682 4.153
8 2930.2 35.57 0.2116 12340.4 3.643 3.673 4.12
9 2904.3 37.83 0.2226 12241.4 3.632 3.664 4.157
11 2923.8 37.20 0.2151 12306.2 3.626 3.655 4.134
12 2902.2 38.43 0.2189 12210.9 3.638 3.657 4.13
13 2890.3 37.83 0.2174 12160.1 3.631 3.663 4.124
14 2951.9 38.87 0.2114 12403.4 3.637 3.664 4.116
15 2908.9 37.20 0.2200 12251.6 3.628 3.662 4.123
16 2962.1 38.03 0.2134 12469.3 3.623 3.654 4.129
17 2927.7 38.23 0.2226 12328.9 3.634 3.679 4.144
18 2873.7 37.60 0.2277 12121.6 3.635 3.672 4.135
19 2926.9 38.83 0.2147 12307.9 3.628 3.661 4.127
20 2926.6 36.60 0.2157 12324.9 3.633 3.676 4.142
21 2912.2 37.43 0.2169 12260.4 3.635 3.678 4.131
22 2915.5 37.63 0.2207 12268.4 3.638 3.671 4.134
23 2911.3 36.57 0.2189 12274.7 3.629 3.663 4.151
24 2933.2 36.17 0.2089 12362.1 3.617 3.651 4.124
25 2890.7 39.20 0.2218 12189.4 3.621 3.658 4.138
26 2915.9 42.44 0.2185 12273.4 3.628 3.658 4.132
27 2915.2 38.72 0.2104 12264.9 3.623 3.654 4.113
28 2923.1 38.52 0.2059 12301.2 3.623 3.666 4.11
29 2907.4 40.09 0.2136 12227.2 3.636 3.676 4.144
30 2909.9 38.43 0.2166 12239.0 3.621 3.668 4.127
31 2979.8 38.50 0.2049 12537.5 3.623 3.659 4.111
32 3016.2 37.13 0.1977 12699.5 3.611 3.647 4.125

4.3. Consistency Analysis Based on Hierarchical Clustering Method

Clustering is an unsupervised knowledge discovery process that groups a set of data such that
the intracluster similarity is maximized and the inter-cluster similarity is minimized [33]. The distance
is a measure of the similarity between samples when the samples are clustered. Assuming that the
“P” is the number of feature quantities for each cell, and the “xij ” represents the j-th feature of the No.
“i” cell, each cell can be viewed as a point in the vector space defined as RP ; therefore, the number of
cell is “n”, and correspondingly the number of point is “n” in RP . A kind of distance should be defined
in RP ; the “dij ” represents the distance between the i-th cell and the j-th cell during the clustering
process, and the closer points tend to be clustered as a cluster or vice versa. The distance commonly
used in clustering is Euclidean distance, which is defined as follows:
" #1/2
p  2
dij = ∑ xik − xjk . (25)
i =1
Appl. Sci. 2017, 7, 802 18 of 21

Due to the different units of cell features, the measurement range varies greatly; therefore, it is
necessary to 7,
Appl. Sci. 2017, normalize
802 data of each feature. The following is the commonly used normalization.
18 of 21

xij − x j
xij0 =' √ xij - x j , i = 1, 2, ..., n, j = 1, 2, ..., p, (26)
xij = sjj ,i = 1, 2,...,n,j = 1,2,...,p , (26)
s jj

2
where the x j = n ∑i=1 xij and the s jj = n−1 ∑in=1 xij − x j is the sample mean and variance of the j-th
1 n 1
where the  = ∑ and the = ∑ ( −  ) is the sample mean and variance of the
feature, respectively.
-th feature, respectively.
The basic idea of hierarchical clustering is that, at first, each cell is considered as a class, and then
The basic idea of hierarchical clustering is that, at first, each cell is considered as a class, and
the nearest distance samples are merged into a new class; at the same time, calculating the distance
then the nearest distance samples are merged into a new class; at the same time, calculating the
between the new class and other classes, repeating two recent classes of mergers, until all cells are
distance between the new class and other classes, repeating two recent classes of mergers, until all
merged into one class. Therefore, the distance between samples before and after vibration could
cells are merged into one class. Therefore, the distance between samples before and after vibration
be calculated. If the sample distance increases, it is shown that cells become dispersed in RP and
could be calculated. If the sample distance increases, it is shown that cells become dispersed in
the
andsimilarity becomes
the similarity worse;
becomes in addition,
worse; in addition, thethemethod
method bybywhich
whichthe theconsistency
consistency deteriorates can be
deteriorates can
further explained.
be further explained.

4.3.1.
4.3.1.Clustering
Clustering Tree
Tree
In
In this
this paper,
paper, the
the average linkagemethod
average linkage methodisisused,
used,asas
thethe space
space is neither
is neither tootoo dense
dense nor nor
too too
expansive. Figure 14 represents the cluster trees of cells before and after vibration.
expansive. Figure 14 represents the cluster trees of cells before and after vibration.

4 32
32 4
31 26
7 31
24 16
12 14
8 24
13 8
6 18
5 28
3 27
2 25
Sample number
Sample number

26 30
21 19
18 15
17 13
25 12
9 11
1 29
16 23
14 9
19 22
29 21
28 20
20 17
11 7
23 1
15 3
30 5
27 6
22 2
1 2 3 4 5 6
1 2 3 4 5 6
Normalized distance Normalized distance

(a) (b)
Figure 14. Clustering tree. (a) Clustering tree before vibration test; (b) Clustering tree after vibration test.
Figure 14. Clustering tree. (a) Clustering tree before vibration test; (b) Clustering tree after
vibration test.
By comparison, when cell samples are classified into a class, the normalized distance of cells
after vibration is greater than before vibration. At the same time, when the standard distance is six,
cell By comparison,
samples when
are divided cell
into onesamples are classified
class before vibration;into a class,after
however, the normalized distance
the vibration, the cell of cells after
samples
vibration
are dividedis greater than
into three before
classes, vibration.
indicating thatAt the
the same time,
consistency of when the
cells are standard distance
deteriorated is six, cell
after vibration.
samples are divided into one class before vibration; however, after the vibration, the cell samples are
4.3.2. Inconsistent
divided Coefficient
into three classes, indicating that the consistency of cells are deteriorated after vibration.
The inconsistent coefficient of the hierarchical clustering method has a self-adaptability feature,
4.3.2. Inconsistent Coefficient
and can show that cells should be divided into several categories according to the characteristics of
the cells’ sample data.coefficient of the hierarchical clustering method has a self-adaptability feature,
The inconsistent
Therefore,
and can show that in cells
this paper,
shouldthe
beinconsistent
divided intocoefficient should beaccording
several categories calculatedtowhen the samples are
the characteristics of the
merged
cells’ intodata.
sample a new class before and after the vibration, respectively; then, the increment of
inconsistent
Therefore,coefficients are obtained.
in this paper, If the increment
the inconsistent is large
coefficient shouldafter the class merging,
be calculated when thethissamples
would are
indicate that the two classes should not be able to merge into a new class, and that the previous
merged into a new class before and after the vibration, respectively; then, the increment of inconsistent class
merging is reasonable. Therefore, the category of the previous class merging as a standard
coefficients are obtained. If the increment is large after the class merging, this would indicate for cellsthat
should be divided into several categories. If results show that there are more categories of cells after
the vibration test, the consistency of the cells which are deteriorated can be concluded.
Appl. Sci. 2017, 7, 802 19 of 21

the two classes should not be able to merge into a new class, and that the previous class merging is
reasonable. Therefore, the category of the previous class merging as a standard for cells should be
divided into several categories. If results show that there are more categories of cells after the vibration
test, the consistency of the cells which are deteriorated can be concluded.
In this paper, the function INCONSISTENT of MATLAB was used to compute the inconsistent
value of each non-leaf node in the hierarchical cluster tree Z, which is a matrix generated by the function
LINKAGE. The Y is a matrix with rows corresponding to each of the non-leaf nodes represented in
Z, therefore the inconsistent value of k-th (k-th indicates the number of batteries classified is k) class
merging is defined as follows.
Z (k, 3) − Y (k, 1)
Y (k, 4) = . (27)
Y (k, 2)
The calculation results of the inconsistent coefficient are presented in Table 6.

Table 6. The inconsistent coefficient of class merging before and after the vibration test.

The Inconsistent Coefficient of Different Classes

Cell Sample Condition
Class 5 Class 4 Class 3 Class 2 Class 1
Pre-Test 2.22 2.51 2.74 3.22 2.95
Post-Test 1.30 2.31 2.60 3.44 3.10

In the case of a cell classification as small as possible, according to Table 6, before the vibration,
the increment on the inconsistent coefficient of the last five times of class merging is 0.29, 0.2, 0.48 and
−0.27, while after the vibration, the increment values on the inconsistent coefficient of the last five
times of class merging are 1.01, 0.29, 0.84 and −0.34. The maximum increment of the inconsistency
coefficient is 0.48 before the vibration, and results show that, when the three classes are merged into
the two classes, the increment of the inconsistent coefficient is the largest; therefore, cells being divided
into three classes before vibration is reasonable. Similarly, cells being divided into five classes is more
reasonable after vibration. Furthermore, it was found that the number of classes of cells after vibration
are increased, which indicate that the consistency of cells is deteriorated.

5. Conclusions
In this paper, the influence of vibration load on electrical performance of lithium-ion cells was
analyzed, the results showed that the mean value and the standard deviation of DC resistance increased
at the 0.05 and 0.1 significance level, respectively. In addition, the significant decrease in cell capacity
and the slight deterioration in consistency was observed after the vibration test.
Compared with previous articles, a large number of samples of 32 lithium-ion cells were used for
the test, and the changes in DC resistances or capacity was compared by using the holistic samples in
order to eliminate the influence of the accidental factors caused by a single cell on the overall trend.
In addition, the distribution function of cell capacity and DC resistance were obtained by data fitting
and using hypothesis testing, which can well reveal the distribution of the cells before and after the
vibration test. Finally, the influence of vibration on the consistency of lithium-ion cells was studied
based on multi-feature quantities, and a reasonable conclusion was obtained by using the hierarchical
clustering method. This point has not been studied in the previous literature, but it must be a research
focus, as the consistency can better reflect the performance of the cell pack.
The quantitative analysis and the error estimation were carried out in the choice of sample size,
and the most representative vibration standard of SAE J2380 was selected after referencing a large
amount of the literature. In addition, the control of the time interval and the ambient temperature were
set reasonably in the measurement process of cell features; therefore, the comprehensive consideration
of the experimental details makes the results more credible.
Appl. Sci. 2017, 7, 802 20 of 21

Acknowledgments: This work was financially supported by the National Key Research and Development
Program of China (No. 2016YFF0203804), the National Natural Science Foundation of China (No. 51375019), and
the Fundamental Research Funds for Central Universities of China (No. FRF-TP-14-061A2).
Author Contributions: Lijun Zhang, Zhansheng Ning and Changyan Sun conceived and designed the
experiments; Zhansheng Ning, Hui Peng and Zhongqiang Mu performed the experiments; Zhansheng Ning and
Hui Peng analyzed the data; and Lijun Zhang and Zhansheng Ning wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wu, L.; Fu, X.; Guan, Y. Review of the remaining useful life prognostics of vehicle lithium-ion batteries using
data-driven methodologies. Appl. Sci. 2016, 6, 166. [CrossRef]
2. Saravanan, M.; Ambalavanan, S. Failure analysis of cast-on-strap in lead-acid battery subjected to vibration.
Eng. Fail. Anal. 2011, 18, 2240–2249. [CrossRef]
3. He, B.; Wang, H.; He, X. Vibration test methods and their experimental research on the performance of the
lead-acid battery. J. Power Sources 2014, 268, 326–330. [CrossRef]
4. Rajalakshmi, N.; Pandian, S.; Dhathathreyan, K.S. Vibration tests on a PEM fuel cell stack usable in
transportation application. Int. J. Hydrogen Energy 2009, 34, 3833–3837. [CrossRef]
5. Betournay, M.C.; Bonnell, G.; Edwardson, E.; Paktunc, D.; Kaufman, A.; Lomma, A.T. The effects of mine
conditions on the performance of a PEM fuel cell. J. Power Sources 2004, 134, 80–87. [CrossRef]
6. Pham, H.L.; Dietz, J.E.; Adams, D.E.; Sharp, N.D. Lithium-ion battery cell health monitoring using vibration
diagnostic test. In Proceedings of the ASME 2013 International Mechanical Engineering Congress and
Exposition, San Diego, CA, USA, 15–21 November 2013.
7. Shevtsov, S.; Chang, S.H. Modeling of vibration energy harvesting system with power PZT stack loaded on
Li-Ion battery. Int. J. Hydrogen Energy 2016, 41, 12618–12625. [CrossRef]
8. Kume, M.; Murakami, T. Battery Array Configured to Prevent Vibration. U.S. Patent 8,999,555, 7 April 2015.
9. Choi, Y.; Jung, D.; Ham, K.; Bae, S. A study on the accelerated vibration endurance tests for battery fixing
bracket in electrically driven vehicles. Procedia Eng. 2011, 10, 851–856. [CrossRef]
10. Reddy, P.R.; Reddy, P.S.; Reddy, K.V.K. Vibration analysis of a Torpedo battery tray using FEA. Int. J. Eng.
Res. Technol. 2014, 03, 128–134.
11. Li, J.; Tian, H.; Wu, P. Analysis of random vibration of power battery box in electric vehicles. In Proceedings
of the 2014 IEEE Conference Transportation Electrification ASIA-Pacific (ITEC Asia-Pacific), Beijing, China,
31 August–3 September 2014; pp. 1–5.
12. Hong, S.K.; Epureanu, B.I.; Castanier, M.P. Parametric reduced-order models of battery pack vibration
including structural variation and prestress effects. J. Power Sources 2014, 261, 101–111. [CrossRef]
13. Arora, S.; Shen, W.; Kapoor, A. Review of mechanical design and strategic placement technique of a robust
battery pack for electric vehicles. Renew. Sustain. Energy Rev. 2016, 60, 1319–1331. [CrossRef]
14. Lee, J.K.; Yeo, J.S.; Jang, M.C.; Yoon, J.M.; Kang, D.M. Mechanical durability and electrical durability of an
aluminium-laminated lithium-ion polymer battery pack for a hybrid electric vehicle. J. Automob. Eng. 2010,
224, 765–773. [CrossRef]
15. Yang, L.; Cheng, X.; Gao, Y.; Yin, G. Effect of vibration on the life of LiCoO2/MCMB battery. Battery Bimon.
2014, 04, 202–205. (In Chinese)
16. An, Y.; Zhang, Z.; Chang, X.; Jiang, T.; Rong, C.; Chen, H. Resarch on vibration resistance of electric vehicle
lithium-ion battery cells. In Proceedings of the China Automotive Engineering Society Annual Meeting,
Beijing, China, 26 November 2013. (In Chinese)
17. Brand, M.; Schuster, S.; Bach, T.; Fleder, E.; Stelz, M.; Glaser, S.; Muller, J.; Sextl, G.; Jossen, A. Effects of
vibrations and shocks on lithium-ion cells. J. Power Sources 2015, 288, 62–69. [CrossRef]
18. Wang, D.; Wu, X.; Wang, Z.; Chen, L. Cracking causing cyclic instability of LiFePO4 cathode material.
J. Power Sources 2005, 140, 125–128. [CrossRef]
19. Hooper, J.; Marco, J. Characterising the in-vehicle vibration inputs to the high voltage battery of an electric
vehicle. J. Power Sources 2014, 245, 510–519. [CrossRef]
20. Hooper, J.; Marco, J. Experimental modal analysis of lithium-ion pouch cells. J. Power Sources 2015, 285,
247–259. [CrossRef]
Appl. Sci. 2017, 7, 802 21 of 21

21. Hooper, J.; Marco, J.; Chouchelamane, G.; Lyness, C. Vibration durability testing of nickel manganese cobalt
oxide (NMC) lithium-ion 18650 battery cells. Energies 2016, 9, 52. [CrossRef]
22. Bruen, T.; Hooper, J.; Marco, J.; Gama, M.; Chouchelamane, G. Analysis of a battery management system
(BMS) control strategy for vibration aged nickel manganese cobalt oxide (NMC) lithium-Iion 18650 battery
cells. Energies 2016, 9, 255. [CrossRef]
23. Hooper, J.; Marco, J.; Chouchelamane, G.; Lyness, C.; Taylor, J. Vibration durability testing of nickel cobalt
aluminum oxide (NCA) lithium-ion 18650 battery cells. Energies 2016, 9, 281. [CrossRef]
24. Marco, J.; Hooper, J. Defining a representative vibration durability test for electric vehicle (EV) rechargeable
energy storage systems (RESS). In Proceedings of the Electric Vehicle Symposium (EVS 29), Montréal, QC,
Canada, 19–21 June 2016.
25. Gao, Y.; Jiang, J.; Zhang, C.; Zhang, W.; Ma, Z.; Jiang, Y. Lithium-ion battery aging mechanisms and life
model under different charging stresses. J. Power Sources 2017, 356, 103–114. [CrossRef]
26. Barai, A.; Chouchelamane, G.H.; Guo, Y.; McGordon, A.; Jennings, P. A study on the impact of lithium-ion
cell relaxation on electrochemical impedance spectroscopy. J. Power Sources 2015, 280, 74–80. [CrossRef]
27. Piłatowicz, G.; Marongiu, A.; Drillkens, J.; Sinhuber, P.; Sauer, D.U. A critical overview of definitions
and determination techniques of the internal resistance using lithium-ion, lead-acid, nickel metal-hydride
batteries and electrochemical double-layer capacitors as examples. J. Power Sources 2015, 296, 365–376.
[CrossRef]
28. Schweiger, H.G.; Obeidi, O.; Komesker, O.; Raschke, A.; Schiemann, M.; Zehner, C.; Birke, P. Comparison
of several methods for determining the internal resistance of lithium ion cells. Sensors 2010, 10, 5604–5625.
[CrossRef] [PubMed]
29. Jossen, A. Fundamentals of battery dynamics. J. Power Sources 2006, 154, 530–538. [CrossRef]
30. Han, X.; Ouyang, M.; Lu, L.; Li, J.; Zheng, Y.; Li, Z. A comparative study of commercial lithium ion battery
cycle life in electrical vehicle: Aging mechanism identification. J. Power Sources 2014, 251, 38–54. [CrossRef]
31. Tang, X.C.; Pan, C.Y.; He, L.P.; Li, L.Q.; Chen, Z.Z. A novel technique based on the ratio of potentio-charge
capacity to galvano-charge capacity (RPG) for determination of the diffusion coefficient of intercalary species
within insertion-host materials: Theories and experiments. Electrochim. Acta 2004, 49, 3113–3119. [CrossRef]
32. Honkura, K.; Takahashi, K.; Horiba, T. Capacity-fading prediction of lithium-ion batteries based on discharge
curves analysis. J. Power Sources 2011, 196, 10141–10147. [CrossRef]
33. Jiang, D.; Pei, J.; Zhang, A. DHC: A density-based hierarchical clustering method for time series gene
expression data. In Proceedings of the Third IEEE Symposium on BioInformatics and BioEngineering,
Hongkong, China, 12–12 March 2003; pp. 393–400.

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).