See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/322284965

Multi-fault online detection method for series-connected battery packs

Conference Paper · October 2017

DOI: 10.1109/CAC.2017.8242769

CITATIONS READS

6 330

5 authors, including:

Yongzhe Kang Bin Duan

Shandong University Shandong University
33 PUBLICATIONS 868 CITATIONS 181 PUBLICATIONS 2,493 CITATIONS

SEE PROFILE SEE PROFILE

Yunlong Shang Zhongkai Zhou

Shandong University Qingdao University
161 PUBLICATIONS 4,307 CITATIONS 26 PUBLICATIONS 927 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Yongzhe Kang on 15 October 2019.

The user has requested enhancement of the downloaded file.

 Multi-fault Online Detection Method for Series-
connected Battery Packs
Yongzhe Kang, Bin Duan, Yunlong Shang, Zhongkai Zhou, Chenghui Zhang*
School of Control Science and Engineering
Shandong University
Jinan, China
zchui@sdu.edu.cn

Abstract—The safety problem is an important factor vehicle complex driving conditions, hence the battery
restricting the development of electric vehicles which caused by connection resistance is gradually increased until a connection
many kinds of faults in the power battery pack. How to diagnose failure occurs [6]. This failure can lead to local temperature
the multiple faults in the battery pack becomes the focus and rise, then a large area thermal runaway, eventually a fire even
difficulty. This paper proposes a multi-fault detection method for explosion. Most of the safety incidents can be avoided or at
battery management systems. Instead of measuring the voltage of least mitigated with proper and reliable management by
individual cells, an interleaved connected voltage measurement detecting initial local faults and handling in a timely manner,
method is presented which considers the effects of the contact and large area thermal runaway can be suppressed
resistance and reliability of voltage sensors. A matrix
prematurely[7-8].
interpretation is developed to demonstrate the viability of the
proposed sensing topology. A methodology is introduced to On the other hand, the battery failure is diagnosed by the
isolate fault type among sensor fault, connection fault and cell voltage, current and temperature signals from sensors. It needs
abuse fault, and then determine where the failure occurs. to be pointed that all sensors have their own reliability, in other
Measurement noise and battery inconsistency have been words, sensor are maybe in fault condition, then the incorrect
considered in MATLAB & Simulink simulation model. acquisition of signals may lead to false diagnosis results,
Simulation results shows the mathematical analysis of resulting in false alarm [9]. Thus, it is critical to distinguish
distinguishing three type faults is correct. The feasibility of among sensor faults, battery abuse faults and connection faults
proposed method is validated.
in order to apply proper mitigation and ensure reliable
Keywords—Power battery; Voltage measurement; Diagnostics;
operation of electric vehicles. An advanced fault diagnosis
Connection fault; Sensor fault system not only can quickly detect the occurrence of failure,
but also accurately determine the type of fault and fault points.
I. INTRODUCTION The most commonly used voltage measurement circuit is
connected to a voltage sensor for each cell, this one-to-one
Lithium-ion batteries are featured by high energy density,
connection type ensures that the voltage of every single cell is
high power density and long life, which have been widely used
monitored [10-11]. But the method is subject to the reliability
in electric vehicles [1]. However, there are still a lot of
of the voltage sensor, the sensor may occur stuck fault (stuck in
potential flaws at the power battery system, bear the brunt of
a certain value), gain fault (scaling fault), offset (bias) fault,
the security problem [2]. A qualified Battery Management
etc. Therefore, the sensor failure cannot be distinguished and
System (BMS) should be able to detect and implement certain
may lead to battery failure error detection.
measures when any of the cell is in fault condition. Advanced
fault diagnosis technology can also be used to ensure the safe Scholars have conducted extensive research on sensor
operation of power battery system in BMS. Meanwhile, there failure problems. One of the effective ways is hardware
are few papers about studying on the failure for the application redundancy, which means the same signal are measured by
of batteries group, and a lack of methods and technologies in multiple sensors at the same time [12], if one of voltage signal
the battery safety issues conspicuously hinder the advance of is abnormal while the two voltage sensor difference is pretty
battery application in EVs [3]. large, you can determine the sensor failure rather than the
battery failure. Another solution is to measure not only the
There are several failure types including sensor fault
voltage of each series unit, but also the total voltage of the
(measurement failure), the battery abuse fault (internal
circuit corresponding to the sum of the series unit voltages to
resistance anomalies, over-voltage, under-voltage,
determine whether a sensor failure occurs. References [13-14]
overcurrent), connection fault, etc. All these failures may be
propose a fault-tolerant voltage measurement method for
evolved into heat failure, burning and even explosion in the
battery management systems. Instead of measuring the voltage
end [4-5]. External or internal short circuit can caused by over-
of individual cells, the proposed method measures the sum
charge/discharge, physical structure damage and other factors.
voltage of multiple battery cells which can distinguish between
In addition to the cell itself, Many modules are connected in
sensor faults and cell faults.
series to constitute a battery string, the module is usually
connected through the copper plate and screws. However, the
connection point may be aged or loose under the electric

978-1-5386-3524-7/17/$31.00 ©2017 IEEE 235

 However all the above methods still exist the same frequent ignition phenomenon at the connection point, as
technical limitation that connection fault and other faults shown in Fig 1(b). There may be a large area of thermal
cannot be distinguished. For example, a connection fault runaway accident eventually. On the other hand, each of the
occurs between two series cells, once the measurement voltage series units in the power packs needs to measure the voltage
is abnormal, the diagnosis may still be mistaken for battery and current, which are passed through the acquisition circuit
abuse fault. However the real point of failure is in the electrode into the BMS to determine the battery State Of Charge(SOC),
connection, not inside the battery, which cannot effectively State Of Health(SOH) and fault state. But it should be noted
identify the real type and location of fault. The problem of that any measurement circuit may be a fault, if a voltage sensor
connection fault can be solved by using the method of
in fault condition, it is prone to lead to a false positive fault
analytical redundancy. Reference [15] use the method of
detection. This may lead to misusage of the protection system,
Shannon entropy and sample entropy to distinguish the fault of
the connection point by distinguishing the battery fault and the resulting in serious consequences.
connection point failure, but these methods need a very large
calculation, which are still difficult to industrial applications.
At present, the most researchers study the battery fault
diagnosis from a certain type of failure, and put forward a
variety of diagnostic methods, however the battery fault types
are diverse, these methods cannot meet all common faults
existing in the power battery pack. A diagnostic method overall
consideration of battery failure which can distinguish all kinds
of faults is necessary. For example, sensor failure and
connection failure can cause abnormal voltage data. There can (a) Power battery pack connection diagram
be misjudgment in distinguishing such similar fault situation,
misjudgment of battery status can cause serious consequences.
In this paper, a multi-fault online detection method for
series-connected battery packs is proposed. Connection fault
and sensor fault and electrical failure in the power battery
system can be effectively diagnosed. The diagnosis principle is
discussed in the paper by matrix analysis. Simulation results
prove that the detection method can isolate the type and
location of a fault robustly by the number of exceptional
voltage sensor.

II. VOLTAGE MEASUREMENT TOPOLOGY DESIGN

(b) Partial enlarged view of the connection fault point
Battery cell will usually appear in a variety of abuse fault,
such as overcharge, over discharge, short circuit, etc. But the Fig. 1. A practical example of a connection failure
power battery system is not only composed of batteries, but
also includes BMS, measuring circuits, connecting accessories, A. Design description
etc. Therefore, its failure also occurs in sensors or connecting Fig. 2 illustrates one embodiment of the proposed voltage
accessories. The battery in the series process is usually measurement method. All resistors Ri ,i +1 represent the
connected with metal pieces, and then tighten the screws, as
shown in Fig 1(a). But in the course of the operation, the connection resistance between the cell i and cell i + 1 . Fig. 3
battery is in the vibrating, temperature-changing and other shows the schematic drawing of the measurement topology. It
complex environments, these connecting accessories may needs to be pointed out that the voltage sensor requires a cross-
connect. The series cells in the battery pack are named as cell 1
appear the phenomenon of aging or loose, so that the contact
resistance between cells increases greatly. Increased contact to cell n. The voltage sensors V1 were connected to the
resistance will further lead to increased heat release, or even negative electrode of cell 1 and negative electrode of cell 2,
V(2n-2) V(2n)

V2 V4 V6 V V

R0,1 Cell1
R1,2 Cell2
R2,3 Cell3
R3,4 Cell(n-1)
Rn −1,n Celln
Rn,n+1

V1 V3 V5 V7 V V

V(2n-3) V(2n-1)

Fig. 2. The proposed voltage measurement method

236
voltage sensors V2 were connected to the external negative voltage of cell i is U ai , Ri ,i +1 represents the connection resistor
output and positive electrode of cell 1; The voltage sensors between the cell i and cell i+1. So the circuit voltage
V2i −1 were connected to the negative electrode of cell i and cell relationship can be expressed as formula (1).
i+1, voltage sensors V2i were connected to the positive The contacting resistance is almost zero in the normal state
electrode of cell i-1 and cell i; At the end, the voltage sensors of the battery, so the sharing voltage can be ignored. However,
V2 n −1 were connected to the negative electrode of cell n and contacting resistance can not be ignored after a connection
external positive output, and voltage sensors V2n were fault occurs in the power pack. It can be seen from equation (1)
that each battery voltage is related to the voltage obtained by
connected to the positive electrode of cell n-1 and cell n. The the two sensors, and each contact resistance is also related to
schematic in Fig. 2 ensures that the voltage of each cell and the voltage obtained by the two sensors.
contact resistance are associated with the measurements of two
sensors. Schematic drawing for cell connections and voltage The difference equation (1) can be written as equation (2):
measurement have been shown in Fig. 3. The topology can
V = AU a - IBR (2)
solve the problem of sensor fault and connect fault. For
example, if the sensor V1 detects an abnormal signal, and the where
associated sensors V2 and V4 are as usual, it can be determined
­V = [V1 ,V2 , ⋅⋅⋅,V2 n ]T
that the sensor V1 is faulty. When cell 3 is in external short °
°
®U a = [U a1 , U a 2 , ⋅⋅⋅,U an ]
T
circuit condition, its terminal voltage drops rapidly, and its (3)
abnormal voltage value will be revealed by V5 and V6. ° T
¯° R = ¬ª R0,1 , R1,2 , R2,3 , ⋅⋅⋅, Rn −1, n , Rn , n +1 ¼º
Similarly, a larger contact resistance can also cause a drop in
the measured voltage, when R2,3 is in connection fault
III. DIAGNOSIS METHOD
condition, abnormal voltage value will be revealed by V3 and
V 6. It should be noted that same battery will not be unlikely to
happen multiple failures at the same time. So all subsequent
analyses are based on ignoring this small probability event.
When the battery voltage measurement sensor failure occurs,
the same battery with two measured values can be compared.
For example, the cell 1 corresponds to the V1 and V2 voltage
sensors when the contact resistance can be ignored. If the two
sensors appear situation that the value of the gap is too large
and the two sides of two sets of related sensors does not appear
abnormal situation, then a battery sensor fault can be detected
and the fault point can be determined by the abnormal sensor
number. Set the sensor fault vector as f s (t ) , and the
reasonable threshold can set as U α , when the V2i −1 sensor be in
Fig. 3. Schematic drawing for cell connections and voltage measurement failure state, fault can be judged by formula (4):
B. Matrix interpretation of measurement topologies f s1 (t ) = V2i −1 − V2i > Uα
Each voltage sensor not only measure the voltage of a
battery cell, but also contains the shared voltage of the near & f s 2 (t ) = V2i +1 − V2i + 2 < Uα (4)
connection resistance. Set the main circuit current as I , the & f s 3 (t ) = V2i −3 − V2i − 2 < Uα
voltage of the voltage sensor is expressed as Vi , The actual

ª V1 º ª1 0 0 0 0 0 ! 0 0º ª0 1 0 0 0 0 ! 0 0º
«V » « » «1
« 2 » «1 0 0 0 0 0 ! 0 0 0 0 0 0 0 ! 0 0 » ª R0,1 º
» ª U a1 º « »« »
« V3 » «0 1 0 0 0 0 ! 0 0» « » «0 0 1 0 0 0 ! 0 0 » « R1,2 »
« » « » « U a2 » « »
« V4 » «0 1 0 0 0 0 ! 0 0» «0 1 0 0 0 0 ! 0 0 » « R2,3 »
« U a3 » « »
« V » = «0 0 1 0 0 0 ! 0 0» « » − I «0 0 0 1 0 0 ! 0 0 » « R3,4 » (1)
« 5 » « »« # » « »
« V6 » «0 0 1 0 0 0 ! 0 0» «0 0 1 0 0 0 ! 0 0» « # »
«U an −1 » « »
« # » «# # # # # # % # #»« » «# # # # # # % # # » « Rn −1,n »
« » « » «¬ U an »¼ « »
«V2 n −1 » «0 0 0 0 0 0 ! 0 1» «0 0 0 0 0 0 ! 0 1 » «¬ Rn,n +1 »¼
« » « » «0
¬ V2 n ¼ ¬0 0 0 0 0 0 ! 0 1¼ ¬ 0 0 0 0 0 ! 1 0 »¼

237
 When a connection fault occurs between the series cells, for represents the connection resistance between each battery. The
example, the contact resistance between the cell i and cell i + 1 normal value of contacting resistance is set to 1mȍ. The
suddenly increases, the voltage of corresponding V2i −1 sensor voltage measurement uses the proposed cross-connect method,
simultaneously the random noise with amplitude of 0.02V is
and V2i + 2 sensor will drop at the same time and the voltage is
added in the voltage output to simulate the actual measurement
lower than the V2i and V2i +1 sensor collect voltage. Set the error. RLC module analog load is connected in series to
contact fault vector as f c (t ) , the fault trigger threshold is I <Δ . represent actual load, and the current sensor is used to measure
So we can determine connection fault by formula (5): the main circuit current.
Normal discharge condition: Battery pack constant load
f c1 (t ) = V2i −1 − V2i discharge, the process did not appear fault situation, all the
= (U ai − IRi ,i +1 ) − (U ai − IRi −1,i ) ( Ri −1,i ≈ 0) voltage data is similar. The results are shown in Fig. 4.

= IRi ,i +1 ≥ I <Δ Normal discharge

& f c 2 (t ) = V2i +1 − V2i + 2 condition

( ) ( )
= U a ( i +1) − IRi ,i +1 − U a ( i +1) − IRi +1,i + 2 ( Ri +1,i + 2 ≈ 0)

Voltage(V)
= IRi ,i +1 ≥ I <Δ
V1 V6
(5) V2 V7
V3 V8
If the battery system is judged by the formula (4) (5) that V4 V9
there are no sensor failure and connection failure in power V5 V10
battery, the battery abuse fault can be further determined. Here,
only a short circuit determination method is listed. The
measurement circuit topology is also suitable for other battery
fault detection. For example, when the cell i is in short circuit Time(s)
condition, the corresponding V2i −1 and V2i voltage sensors will
drop at the same time. The method of distinguishing between Fig. 4. Normal discharge condition simulation.
connecting fault is the difference of abnormal sensor number.
The method of distinguishing between sensor fault is the
number of abnormal sensor. It is possible to determine short
circuit in cell i whether the absolute voltage difference of
V2i −1 and V2i is less than the threshold U σ .

f s1 (t ) = V2i −1 − V2i
Voltage(V)

= (U ai − IRi ,i +1 ) − (U ai − IRi −1,i ) ( Ri −1,i ≈ Ri −1,i ≈ 0)

= Uδ 1 − U δ 2 ≤ Uσ
& f s 2 (Δt ) = V2i −1 (t0 ) − V2i −1 (t0 + Δt ) ≥ Uη V5 voltage sensor
data
& f s 3 (Δt ) = V2i (t0 ) − V2i (t0 + Δt ) ≥ Uη
˄6 ˅
For example, set the contact fault vector as f s (t ) , the
threshold of voltage change in Δt is Uη , The voltage Time(s)
measurement error is expressed as U δ . So we can determine Fig. 5. Sensor fault simulation for V5 at 1 s.
short circuit fault in cell i by formula (6).
Sensor fault: It should be noted that the sensor failure can
be divided into stuck state, constant gain fault, constant
IV. SIMULATION RESULTS AND ANALYSIS
deviation fault and so on. In this paper, only a constant
The simulation is set up to test the viability of the multi- deviation fault is listed, and other faults can still be diagnosed
fault online detection method. A SIMSCPAPE model is built in by a similar method. An Sensor fault on V5 is induced at 1s by
MATLAB & Simulink for a battery pack consisting of five reducing by 2 V, the reasonable threshold U α can set as 0.5V.
battery cells in series connection. The battery model uses a It can be seen in Fig. 5 that V5 sensor data voltage drop at
lithium-ion battery model, sets an average cell voltage of 7.2V, t = 1s , meanwhile the rest of the sensor are normal. Therefore
and sets a maximum 8% difference of the SOC initial value to
V5 sensor fault can be judged through the fault diagnosis
simulate the inconsistency between the series cells. The
principle as shown in formula (7).
resistance is connected in series between cells, which

238
 f s1 (t ) = V5 − V6 > Uα connection fault, therefore we can determine by f s 2 (Δt ) and
f s 3 (Δt ) that the cell 3 is in short-circuit fault.
& f s 2 (t ) = V7 − V8 < Uα (7)
& f s 3 (t ) = V3 − V4 < Uα f s1 (t ) = V5 − V6
Connection fault: A connection fault between cell 2 and = (U a 3 − IR3,4 ) − (U a 3 − IR2,3 ) ( R2,3 ≈ R3,4 ≈ 0)
cell 3 is induced at 1s by suddenly increasing the resistance to
0.5ȍ. The fault trigger threshold I <Δ can set as 1V. The = Uδ 1 − U δ 2 ≤ Uσ (9)
results are shown in Fig. 6. V4 and V5 voltage sensor voltage & f s 2 (Δt ) = V5 (t0 ) − V5 (t0 + Δt ) ≥ Uη
drop at t=1s, but they measure the voltage of cell 2 and cell 3
respectively. & f s 3 (Δt ) = V6 (t0 ) − V6 (t0 + Δt ) ≥ Uη
Voltage(V)

V4 sensor Voltage(V)
V5 sensor V5 sensor
V6 sensor

Time(s)
Fig. 6. Connection fault between cell 2 and cell 3 at 1 s. Time(s)
By the determination principle, the voltage of
Fig. 7 Short circuit fault in cell 3 at 0.67s.
corresponding V2i −1 sensor and V2i + 2 sensor will drop at the
same time and the voltage is lower than the V2i and
V. CONCLUSION
V2i +1 sensor collect voltage. It can be determined by formula (8)
that the contact resistance between the cell 2 and cell 3 is too This paper proposes a multi-fault online detection method
large, there have been connection fault. for series-connected battery packs to distinguish among sensor
fault, the battery abuse fault and connection fault. First, we
f c1 (t ) = V3 − V4 present a voltage measurement method which considers the
effect of contact resistance and reliability of voltage sensor.
= (U a 2 − IR2,3 ) − (U a 2 − IR1,2 ) ( R1,2 ≈ 0) Each voltage sensor by interleaved connected measure not only
the voltage of a battery cell, but also the shared voltage of the
= IR2,3 ≥ I < Δ near connection resistance. The matrix analysis shows the
(8)
& f c 2 (t ) = V5 − V6 viability of the proposed measurement topology in isolating
abuse fault, sensor failure and connection fault. Each battery
= (U a 3 − IR2,3 ) − (U a 3 − IR3,4 ) ( R3,4 ≈ 0) voltage and contact resistance voltage are associated with two
sensors. After voltage measurement method design and a
= IR2,3 ≥ I < Δ preliminary analysis, a SIMSCPAPE model is built in
Short circuit fault: It should be explained that this article MATLAB & Simulink for a battery pack consisting of five
only lists a simple way to determine the short circuit, in order battery cells in series connection. The viability of the multi-
to explain how to apply the universal method to the proposed fault online detection method is proved by simulation results.
topology, the method of determining battery abuse fault is not
analyzed in detail. A short circuit fault is induced at 1s by ACKNOWLEDGMENT
suddenly short-circuit the cell 3 with 0.5ȍ parallel resistance at
This project is supported by the National Natural Science
t=0.67s. U δ is set to 0.5V, and Uη is set to 1V when Δt is Foundation of China (No. 61527809) and (No.61633015)and
0.1s. The results are shown in Fig. 7, the voltage of V5 and V6 Key research and development program of Shandong Province
voltage sensor appear drop situation at the same time. These (2016ZDJS03A02) and Shandong University young scholars'
two sensor are used to measure the voltage of the cell 3. The future plan (2017WLJH26).
f s1 (t ) determines that cell 3 does not have a sensor failure or

239
 REFERENCES [9] S. Dey , S. Mohon , P. Pisu , etc. “Sensor Fault Detection, Isolation, and
Estimation in Lithium-Ion Batteries,” IEEE Transactions on Control
[1] L. Lu, X. Han, and J. Li, etc. “A review on the key issues for lithium-ion Systems Technology, vol. 24, pp. 2141-2149, 2016.
battery management in electric vehicles,” Journal of Power Sources, vol.
226, pp.272-288, 2013. [10] Y. Shang , C. Zhang , and N. Cui, “A Cell-to-Cell Battery Equalizer
With Zero-Current Switching and Zero-Voltage Gap Based on Quasi-
[2] C. Wu, C. Zhu, and Y. Ge, etc. “A Review on Fault Mechanism and Resonant LC Converter and Boost Converter,” IEEE Transactions on
Diagnosis Approach for Li-Ion Batteries,” Journal of Nanomaterials, Power Electronics, vol. 30, pp. 3731-3747, 2015.
vol. 2015, pp.1-9, 2015.
[11] Y. Shang , C. Zhang , and N. Cui, “A Crossed Pack-to-Cell Equalizer
[3] G. Kim, K. Smith, and J. Ireland, etc. “Fail-safe design for large Based on Quasi-Resonant LC Converter with Adaptive Fuzzy Logic
capacity lithium-ion battery systems,” Journal of Power Sources, vol. Equalization Control for Series-connected Lithium-Ion Battery Strings,”
210, pp.234-253, 2012. Applied Power Electronics Conference & Exposition, pp.1685-1692,
[4] Y. Zheng, X. Han, and L. Lu, etc. “Lithium ion battery pack power fade 2015.
fault identification based on Shannon entropy in electric vehicles,” [12] V.P. Nelson, “Fault-Tolerant Computing: Fundamental Concepts,”
Journal of Power Sources, vol. 223, pp.136-146, 2013. Computer, vol. 23, pp. 19-25, 1990.
[5] Y. Song, Y. Yang, Z. Hu, “Present status and development trend of [13] B. Xia , C. Mi, “A fault-tolerant voltage measurement method for series
batteries for electric vehicles,” Power System Technology, vol. 35, pp.1- connected battery packs,” Journal of Power Sources, vol. 308, pp. 83-96,
7, 2011. 2016.
[6] K. Daeil, Y. Jeongah, “A model-based prognostic approach to predict [14] B. Xia, T. Nguyen , J. Yang , etc. “The improved interleaved voltage
interconnect failure using impedance analysis,” Journal of Mechanical measurement method for series connected battery packs,” Journal of
Science and Technology, vol. 30, pp. 4447-4452, 2016. Power Sources, vol. 334, pp. 12-22, 2016.
[7] N. Williard, W. He, C. Hendricks, etc. “Lessons Learned from the 787 [15] L. Yao , Z. Wang , J. Ma, “Fault detection of the connection of lithium-
Dreamliner Issue on Lithium-Ion Battery Reliability,” Energies, vol. 6, ion power batteries based on entropy for electric vehicles,” Journal of
pp. 4682-4695, 2013. Power Sources, vol. 293, pp. 548-561, 2015.
[8] Q. Wang , P. Ping , X. Zhao ,etc. “Thermal runaway caused fire and
explosion of lithium ion battery,” Cheminform , vol. 43, pp. 210-224,
2012.

240

View publication stats