1 Design and Development of a Power Battery Insulation Fault Diagnosis System

2 Based on Fuzzy Theory

ed
3 Kai Wen , Xingfeng Fu3 , Feng Pei2，Hai Jiang1
1,2

4 （[Link] University, Beijing, China. 2. Guangzhou Greater Bay Technology Co., Ltd, Guangzhou
5 China. 3. GAC AION NEW ENERGY AUTOMOBILE CO.,LTD Guangzhou China.)
6 Corresponding author: Xingfeng Fu. E-mail: fxf1000@[Link]

iew
7 Abstract: In depth exploration and research have been conducted on the identification and judgment of
8 power battery system faults. A power battery fault diagnosis and service system based on fuzzy theory
9 has been constructed. By analyzing the classification of power battery faults and the collected fault signal
10 framework samples, fuzzy theory is used to analyze and identify battery system faults. Through
11 simulation and experimental comparison, it can be concluded that, The fuzzy recognition system for

v
12 power battery faults constructed has good recognition accuracy and short recognition time, and has good
13 engineering promotion value.

re
14 Keyword: Battery fault diagnosis system, Insulation, Fuzzy theory, Fault identification.
15
Abbreviations
16
17
EV electric vehicle DEM Diagnostic event manager
18
19
DTC
DID
Diagnostic Trouble Code
Date Identifier
er BSM
CDD
Basic software
CANdela Diagnostic Descriptions
20
ODX Open Diagnostic Data Exchange SWC Software component
21
pe
ECUM ECU state manager FIM Function inhibition manager
22
DCM Diagnostic communication manager NVM NVRAM manager
23
RTE Run-Time Environment EDV Emergency vehicle driver
24
ERA Emergency Response Action IUD Integration unit driver
25
PCM phase change material PTC positive temperature coefficient
26
ot

EDS electric drive system BMS battery management system

27
PRA power relay assembly IPU integrated power unit
28
AC air conditioning VCU vehicle control unit
29
tn

IDI Integrated Discrimination Improvement

30

31 1. Introduction
32 Power batteries are the core components of electric vehicles, and battery management
33 systems(BMS，Battery Management System) and application technologies are the diagnostic and
rin

34 control technologies of power batteries. One of the core contents of power battery system development
35 is the ability to timely and accurately detect, identify, and judge battery faults, which can effectively
36 extend the service life of power batteries, improve their consistency and reliability. However, during
37 the use of power batteries, there is complexity and ambiguity in the causes of faults, and misjudgment
ep

38 should not affect the normal use of power batteries. [1] Therefore, it is necessary to achieve optimal
39 judgment in power battery fault diagnosis and prevention of misjudgment to ensure the safety and
40 convenience of electric vehicle use.[2]
41 At present, there are two diagnostic communication protocols used by various vehicle
Pr

42 manufacturers around the world: KWP2000 and UDS (Unified Diagnostic Services). We have chosen
43 the more popular UDS diagnostic communication protocol among them. The UDS diagnostic
44 communication protocol consists of a data link layer, a network layer, and an application layer. The data

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 45 link layer is mainly responsible for implementing underlying CAN (Controller Area Network) driver
46 functions, the network layer implements multi frame transmission protocols, and the application layer

ed
47 implements specific diagnostic services and corresponding communication protocols. The fault handling
48 function for specific faults can be roughly divided into several parts, including fault detection, fault
49 management, and system fault response and handling. [3-4]
50 The application environment of power batteries in electric vehicles is very complex, and there are

iew
51 various interference factors. Therefore, it is very important to timely and accurately judge the fault
52 information. Only by accurately judging various fault information can have targeted and effective
53 measures be taken to reduce fault injuries. To solve this problem, it is necessary to construct a diagnostic
54 model for battery faults based on existing battery big data fault information, combined with high-
55 precision mathematical models obtained from previous experiments. By combining various sensor

v
56 signals and interference information in the vehicle, key judgment factors can be correctly identified, fault
57 information can be located, and effective processing measures can be taken to achieve fault isolation.

re
58 2. BMS Development of diagnostic functions
59 The main goal of developing BMS application software is to monitor battery information and ensure
60 the normal use of the battery system; At the same time, on the basis of strengthening management, strive
61 to improve battery efficiency as much as possible and ensure safe use. [5-6]
er
62 The UDS diagnostic protocol is a requirement specification for diagnostic communication in the
63 automotive industry, defined by the ISO-14229 series of standards. The application layer (Layer 7)
64 applied to the OSI seven-layer model only specifies service requirements related to diagnosis and does
pe
65 not involve communication mechanisms. Therefore, it can be implemented on different automotive buses
66 (such as CAN, LIN, Flexray, Ethernet, and K-line). [7-8]
67 Diagnostic service is an information exchange method between diagnostic equipment and ECU
68 (Electronic Control Unit) [9] Usually, a diagnostic device sends a request and the ECU responds. A fault
69 code is a code used to mark ECU faults, which follows certain rules and is stored in the non-volatile
ot

70 storage of the ECU. Diagnostic data is the internal data of the ECU that can be requested by diagnostic
71 equipment, including: current data: data that the ECU is running, such as vehicle speed, throttle opening,
72 engine speed, etc; Storage data: The data stored in the memory by the ECU at a certain time, such as
tn

73 DTC; Static data: Constant and unchanging ECU internal data, such as VIN codes. [10]
rin
ep
Pr

74
75 Fig.1 Definition of Application Layer Timing Parameters for UDS Diagnostic Protocol

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 76 The implementation phase of BMS fault diagnosis software includes implementing fault detection
77 function in TargetLink model, fault response and processing function in TargetLink model, DTC

ed
78 (Diagnostic Trouble Code) storage function in C code, fault related diagnostic service function in C code,
79 and software integration from UDS protocol stack code to application layer C code and TargetLink model
80 in C code.[11] The specific process is showed in Fig 2.
Implementing Implement

iew
Implementing Implement software
fault response Implement fault related
fault detection integration from UDS
and handling DTC storage diagnostic
function in the protocol stack code to
functions in the function in C service
TargetLink application layer C code
TargetLink code functions in
model and TargetLink model in C
model C code
81 code
82 Fig.2 BMS software fault diagnosis design and development process

v
83 The testing and verification phase is to verify whether the BMS system fault diagnosis strategy
84 software implementation meets the UDS protocol specifications and fault diagnosis strategy

re
85 specifications, mainly including UDS protocol CAN delta automatic testing, fault DTC detection and
86 storage HIL testing, fault response HIL testing, joint testing with on-board diagnostic equipment, and
87 joint testing with offline testing equipment.
UDS
Diagnostic Joint testing
Protocol
CANdela
Fault DTC
detection and
storage HIL test
er Fault
handling HIL
testing
of onboard
diagnostic
Offline detector joint
testing
automatic equipment
testing
88
pe
89 Fig.3 Diagnostic service testing process

90 3. A Fuzzy System Based Algorithm Model for Battery Fault Diagnosis

91 The electrochemical model of lithium-ion battery is an electrochemical equation model that can
92 characterize the characteristics of lithium-ion battery through comprehensive analysis of the internal
ot

93 working mechanism of lithium-ion battery, the material of lithium-ion battery, the spectrum of lithium-
94 ion battery and the internal diffusion effect. [12-15]

95 {𝑈𝑑𝑈= 𝑈𝑑𝑡 =(𝑆𝑂𝐶)

𝐿 ― 𝑖(𝑡)𝑅 ― 𝑈
𝑂𝐶
―𝑈 𝑅 𝐶 +𝑖 𝐶
𝑂 𝑃
(1)
tn

𝑃 𝑃 𝑃 𝑃 𝑃

96 𝑈𝑃𝑏 = ∫ 𝑖𝑑𝑡 + 𝑈𝑃𝑏0 (2)

97 In the formula, UL is the terminal voltage, Uoc is the open-circuit voltage, i (t) is the load current,
98 R0 is the ohmic internal resistance, Cp is the equivalent capacitance of the electrode plate, Up is the
rin

99 terminal voltage at both ends of the polarization capacitor Cp, i is the current, Cb is the equivalent
100 capacitance describing the open-circuit voltage change, and Upb is the terminal voltage at both ends of
101 the capacitor Cb.

⌈ ⌉
∆𝑡

{
―𝜂
⌈ 𝑆𝑂𝐶𝑘
⌉=[ 1 0
][ 𝑆𝑂𝐶𝑘―1
]+ 𝐶𝑚𝑎𝑥
𝑖𝑘 + 𝑤
ep

𝑈𝑝,𝑘 0 ―∆𝑡 𝑈𝑝,𝑘―1 ―∆𝑡

102 𝑒 𝜏 𝑅𝑝(1 ― 𝑒 𝜏 ) (3)
𝑈𝐿,𝑘 = 𝑈𝑜𝑐(𝑆𝑂𝐶𝑘) ― 𝑈𝑝,𝑘 ― 𝑖𝑘𝑅𝑜 + 𝜈

103 In the formula, Cmax is the maximum usable capacity of the battery, η is the coulomb efficiency,
Pr

104 ∆ t is the sampling period, τ = 𝑅𝑃𝐶𝑃is the time constant, w is the process noise, and v is the
105 observation noise.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 𝜂𝑖Δ𝑡𝑖(𝑘)
106 SOC(𝑘 + 1) = SOC(𝑘) ― (1 ― 𝜂𝑛)𝐶𝑛 (4)
𝜂𝑇

ed
107 In the formula, SOC (k+1) is the SOC value of the battery at the time of k+1; I (t) is the current
108 value at the time of k~k+1; Qn is the rated capacity of the battery; η i， η T and η N represents the
109 correction coefficient of the charge-discharge ratio, temperature and charge-discharge cycle number on
110 the battery capacity, which can be obtained by processing the data of the charge-discharge experiment

iew
111 of the battery.[16]
1

{ [] ( [ ]
0 0

112
𝑈𝑏
𝑈𝑝
= 0 ―1 𝐶 𝑅
𝑝 𝑝
)[ 𝑈𝑏
]
𝑈𝑝 + 1
𝐶𝑏
𝐶𝑝
𝐼𝐿
(5)
[𝑈𝐿] = [ ―1 [ ]
𝑈
―1] 𝑈𝑏 + [ ― 𝑅𝑂][𝐼𝐿] + [𝑈𝑂𝐶]
𝑝

v
113 In the model, the voltage Ub and Up is the state variable of the two capacitors, and the terminal

re
114 voltage UL of the battery is the output variable.
115 Fuzzy neural networks are neural networks with fuzzy weight coefficients or input signals that are
116 fuzzy quantities. Fuzzy neural networks, whether used as approximators or pattern memory, require
117 learning and optimizing weight coefficients. Learning algorithms are the key to optimizing weight
118 coefficients in fuzzy neural networks. For logical fuzzy neural networks, error based learning algorithms,
119
er
also known as supervised learning algorithms, can be used. For arithmetic fuzzy neural networks, there
120 are fuzzy BP algorithm, genetic algorithm, etc. There is currently no reasonable algorithm for hybrid
121 fuzzy neural networks; However, hybrid fuzzy neural networks are generally used for computation rather
pe
122 than learning, and they do not necessarily need to be learned. [17]
123 The most effective and widespread application of fuzzy theory is in fuzzy control, which
124 unexpectedly solves problems that traditional control theory cannot solve or is difficult to solve in various
125 fields, and has achieved some convincing results. Summarize the control strategies of human experts for
126 specific controlled objects or processes into a series of control rules expressed in the form of "IF
ot

127 (condition) THEN (action)", obtain the control action set through fuzzy reasoning, and apply it to the
128 controlled object or process. The control action set is a set of conditional statements, while the state
129 statement and control action are both quantified fuzzy language sets. [18]
tn
rin
ep

130
131 Fig.4 The relationship between the number of iterations, Spread membership, crossover probability, and
Pr

132 mutation probability membership functions of fuzzy functions

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
iew
133

v
134 Fig.5 Fuzzy System Output Diagram of Cross Probability

re
135 The diversity, convergence, and number of iterations of the sample population are closely related
136 to the performance indicators of the algorithm and the algorithm results. Therefore, the number of
137 iterations and convergence indicators are used as inputs to the fuzzy system, and the crossover probability
138 and mutation probability are dynamically adjusted during each iteration process. In order to measure the
139 number of iterations of the algorithm, the iteration value is limited to the interval [0,1], and the
140
er
convergence metric spread is normalized. The input of each sample is granulated into three triangular
141 membership functions (low, mid, and high respectively represent the size of the input value), and the
142 output is granulated into five triangular membership functions (low, mid low, mid, mid high, high
pe
143 respectively represent the size of the output value)。
ot
tn
rin
ep

144
145 Fig.6 Fault handling flowchart based on Fuzzy System
Pr

146 Fig 6 is a fault diagnosis and judgment flowchart based on the application of fuzzy theory. The fault
147 diagnosis of the vehicle controller mainly includes fault detection, fault management, and fault handling.
148 The specific diagnostic process is shown in Figure 6. Firstly, based on the VCU input/output interface

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 149 and functional definition, fault detection is carried out to confirm the specific faults currently present in
150 the VCU; Then, for confirmed faults, they will be automatically stored at the bottom layer based on pre-

ed
151 defined fault codes and frozen frames; Finally, different faults will directly correspond to corresponding
152 fault handling actions, and each functional module will then perform specific failure protection actions.
153 As shown in Fig6, DFC file is diagnostic fault check file while FID is Fault Identification
154 Denotation. After receiving various fault information, the processor will make judgments and analyze

iew
155 processing in the fault detection module, form corresponding menu files, and predict the next step of
156 fault management information by comparing the corresponding fault code information based on the FRM
157 model. In the fault management information module, based on the judgment logic and calibration
158 parameters constructed from the corresponding DTC and DID lists, corresponding fault levels and
159 expected fault actions are divided, In the fault handling module, the driver's behavior and vehicle status

v
160 will be comprehensively judged based on the obtained fault management information and corresponding
161 fault levels, and corresponding fault actions will be executed based on various types of information. The

re
162 entire fault diagnosis and handling process is constantly being repeated and confirmed based on changes
163 in driving conditions.

164 4. Identification and isolation of Battery Faults

165 The signals and communication information collected from electric vehicles are mixed with a large
er
166 amount of error noise. In order to reduce the probability of false alarms and improve detection accuracy,
167 it is necessary to set some judgment rules to determine whether corresponding faults have actually
168 occurred. In this study, the collected CAN communication signals and sensor fault information are used
pe
169 as input samples for the fuzzy system, and corresponding estimation thresholds are formulated, To
170 analyze and judge the corresponding fault information, and finally confirm whether the fault has actually
171 occurred. Only after confirmation can the corresponding fault handling measures be implemented. [19]
172 After BMS detects a fault, it will start an appropriate debounce process. Both fault confirmation
173 and fault recovery must go through a calibrated debounce time. The Debounce module is a timer that
ot

174 starts running when the enable signal MonEna is 1, and the program can control its reset. After reset or
175 initialization, the timer value will be set to 0. [20]
176 If the monitoring result MonDfct of the fault detection report is 1, it indicates that a fault has
tn

177 occurred and the "fail" timer increases. Once this timer reaches the calibratable threshold of "fail" (tiDfct),
178 bit 0, bit 1, and bit 3 of the status word stDbc will be set to 1.
179 If the monitoring result MonDfct of the fault detection report is 0, that is, no fault has occurred, the
180 "pass" timer will increase. Once this timer reaches the calibrated threshold (tiOk) for "pass", the bit 0 of
181 the status word stDbc will be reset to 0.
rin

182 The anti shake debounce module of HCU follows the following rules: the debounce module is
183 designed based on a counter, which increases when the reported fault monitoring result MonOk is 1. If
184 the reported fault monitoring result is 0, it indicates that no fault has been detected, and this counter
185 decreases.
ep

186 Due to this software structure, fault detection "pass" and "fail" can never occur simultaneously.
187 Tab 1. Main battery failure threshold table
Fault Clear vehicle operation Vehicle standing still
Diagnostic Fault
Pr

Fault Fault toler Diagnos fault Recovery External

Item start recovery Dsichar Dsichar
name threshold ance tic cycle conditio time Charge charging Charg
conditions value ge ge
time ns e

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 >4.20V 4s 10ms ≤4.18V 2s Map 0 Stop Stop enable

ed
cel
>4.25V 4s 10ms ≤4.2V 2s 10kW 0 cut off Stop enable

voltage
unable
From
too high
>4.30V 1s 10ms to / Power down cut off Stop Stop
initializatio
Automat
recover
cell n to low

iew
ic
≤2.8V 4s 10ms >2.9V 2s 10kW Map enable enable enable
voltage voltage
clearing
Request to power
power
cell ≤2.5V 2s 10ms >2.8V 2s enable enable Stop
down
down
voltage
unable
too low
≤2.0V 1s 10ms to / Power down Stop Stop Stop

recover

v
Module ≥50℃ 4s 10ms ≤45℃ 4s Map Map enable enable enable

re
temperatu Request to power
≥55℃ 4s 10ms ≤50℃ 4s Stop Stop Stop
re too down
Module
high ≥60℃ 4s 10ms ≤55℃ 4s Power down Stop Stop Stop
temperatu / /
≤-20℃ 4s 10ms ≥-17℃ 4s Map Map Stop Stop enable
re Module
≤-25℃ 4s 10ms ≥-22℃ 4s Map Map Stop Stop enable
temperatu

re too low ≤-30℃ 4s

er
10ms ≥-27℃ 4s
Request to power
Stop Stop Stop
down

<0.95*
pe
From
>0.9*DCH DCH immediatel
discharge 2s initializatio 10ms 10kW Map / enable enable
threshold threshol y
current n to low
d
overcurre voltage
<DCH
nt ≥DCH power Request to power
2s 10ms / threshol 100ms / enable enable
ot

threshold down down

d

Short

circuit of 500 Request to power

tn

≥550A 10ms <550A 100ms / enable enable

discharge ms down
From
current
current initializatio
<0.95*R
n to high
>0.9*RCH CH immediatel Request to power
rin

2s voltage 10ms / enable enable

Recharge threshold threshol y down
power
overcurre d
down
nt <RCH
≥RCH Request to power
2s 10ms / threshol 100ms / enable enable
ep

threshold down
d

From <0.95*C
Charging
>0.9*CH initializatio H immediatel
overcurre 2s 10ms / / / enable enable
threshold n to stop threshol y
Pr

nt
charging d

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 <CH
≥CH
2s 10ms threshol 100ms / / / /

ed
threshold
d

≥1250Ω
≤1000Ω/V 20s 10ms 20s / / enable enable enable
/V

AC ≥1000Ω

iew
≤750Ω/V 20s 10ms 20s / / enable enable enable
insulation From /V

initializatio ≥550Ω/
≤300Ω/V 20s 10ms 20s / / Stop Stop Stop
n to high V

voltage ≥1250Ω
≤1000Ω/V 20s 10ms 20s Map Map / enable enable
DC power /V

v
insulation down ≥1000Ω
insulation ≤750Ω/V 20s 10ms / 20s 10kW 10kW / enable enable
(closed /V

re
relay) ≥750Ω/ Request to power
≤500Ω/V 20s 10ms 20s / Stop Stop
V down

From ≥1250Ω
≤1000Ω/V 20s 10ms 20s Map Map / enable enable
DC initializatio /V

insulation

(open
≤750Ω/V 20s
n to low

voltage
er
10ms
≥1000Ω

/V
20s 10kW 10kW / enable enable

relay) power ≥750Ω/ Request to power

≤500Ω/V 20s 10ms 20s / Stop Stop
pe
down V down

188 Table 1 is the compiled threshold table for power battery faults, which is used to determine whether
189 a fault has occurred by combining the control thresholds in Table 1 with relevant auxiliary signals. Taking
190 the insulation fault diagnosis in the table as an example, after detecting a signal with a low insulation
191 resistance value, due to the CAN bus communication cycle of the insulation fault being 1 second, the
192 BMS needs to receive a signal with a low insulation resistance value for 20 consecutive times, add other
ot

193 auxiliary judgment information, and comprehensively determine the insulation fault before executing the
194 corresponding reporting and processing measures. For example, if the insulation resistance value is ≤
195 1000 Ω/V, the BMS will report a level one minor insulation fault information, and the vehicle can still
tn

196 be used normally. The charging and discharging power map remains unchanged. When the insulation
197 resistance value is ≥ 1250 Ω/V, the fault will automatically recover; When the insulation resistance
198 value is ≤ 750 Ω /V, the BMS will report a secondary insulation fault information. The battery
199 discharge power and feedback power are both limited to 10kW, and the whole vehicle enters a limp state.
rin

200 At this time, the charging and discharging power of the battery is very small, and the discharge power is
201 only enough for moving the vehicle. When the insulation resistance value is ≥ 1000 Ω/V, the fault
202 will automatically recover; When the insulation resistance value is ≤ 500 Ω/V, the BMS will report
203 a three-level insulation fault information. The battery discharge power and feedback power are limited
ep

204 and cannot be used, and the entire vehicle is prohibited from running. The BMS requests to power off.
205 When the insulation resistance value is ≥ 750 Ω/V, the fault will automatically recover.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
v iew
re
206
207 Fig.7 Insulation resistance simulation circuit diagram ( DC side)
208 Fig 7 is an insulation simulation model diagram of the high-voltage battery part in the power battery
209 high-voltage circuit system established in Simulink software. In this simulation model, the BMS is first
210
er
awakened to perform the power on operation. After the BMS is powered on, at 3 seconds, a connection
211 resistance of 100k Ω is introduced to determine whether the entire insulation detection system can detect
212 insulation faults inside the battery pack in a timely manner, And whether the fault handling logic executed
pe
213 is consistent with the corresponding fault control threshold, and whether the processing flow of the
214 actuator conforms to the designed fault handling flow. Does the time for fault diagnosis and response
215 time meet the design requirement?
216 Fig 8 shows the results obtained from simulation calculations. From the results in Figure 8, it can
217 be seen that after the BMS wakes up and starts normally, the self check passes, and the BMS state value
ot

218 is referred to as 2 (main relay closed). The BMS enters the working state, and after calibration at the 3
219 second position, a resistor with a series resistance value of 100k Ω is introduced. After about 3 seconds
220 of judgment time, the BMS detects an insulation fault and reports the insulation fault status. In a relatively
tn

221 short time of about 150ms, BMS reports a fault, and the status value of the BMS is set to 9 (power hold
222 relay open) in a short period of time. After repeated confirmation for about 100ms, the BMS reports an
223 insulation fault, and the entire simulation result meets the design requirements of the control threshold
224 table. The response time meets the design requirements. Therefore, the simulation results prove that after
rin

225 introducing relevant series resistors, the BMS can make accurate fault diagnosis results and execute
226 actions.
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
iew
227

v
228 Fig.8 Simulation result of battery Insulation resistance based on Fuzzy System

229 5. Discussion on Experiment results

re
er
pe
ot

230
tn

231 Fig.9 BMS fault test and insulation fault test platform
232 Fig 9 is an experimental setup diagram of the experimental setup. Build the corresponding test
233 bench according to the testing requirements. The upper left image in the figure shows the HIL detection
234 equipment, which simulates various faults and communication signals of the power battery system
rin

235 through the HIL experimental equipment to verify the recognition rate of the constructed fuzzy judgment
236 system. The number of test samples in this experiment is set to 100, which is also close to the number of
237 fault samples tested in actual vehicles. Therefore, the results of this setting are more representative. The
238 upper middle position of Figure 6 is an experimental device that simulates various input and
ep

239 communication signal errors and interference signals of the BMS. The accuracy of the BMS model
240 judgment is analyzed and judged through the simulated signals. The lower left corner of Figure 6 shows
241 the testing bench where the actual measurement signal error of the battery pack is introduced. By
242 introducing measurement errors from various sensors, the accuracy of fault diagnosis in the simulation
Pr

243 system is analyzed and tested. The right part of Figure 9 is a device for measuring insulation faults in
244 real vehicles, which determines the accuracy of insulation faults by introducing insulation values of
245 various resistance values.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 246 Fig 10 shows the recognition and calculation results of insulation faults and overcurrent faults in
247 the fault signal of the constructed fuzzy system. In the first 50 seconds, no faults are input. After 50

ed
248 seconds, after introducing fault signals and interference signals, and separately introducing insulation
249 fault signals and overcurrent fault signals, it can be seen from Figure 7 that in the first 50 seconds without
250 faults, the fault model values and the system's judgment values are very consistent, without residual
251 values; After introducing simulated fault signals and interference signals, there was a deviation between

iew
252 the system's judgment value and input fault value, forming a certain residual value. However, the residual
253 values were very small, both within 1.0. After about 220 seconds, the insulation fault residual value
254 approached 0, and the model error and judgment value error basically disappeared; After about 330
255 seconds, the overcurrent error value and judgment value error basically disappear. After this, the fuzzy
256 recognition system can basically recognize the corresponding fault signal content. After repeated testing,

v
257 the accuracy of the model's judgment error is about 92.7%, which is an ideal result.

re
er
258
pe
259 Fig.10 Residual of insulation fault model based on Fuzzy System
ot
tn

260
261 Fig.11 Schematic diagram of insulation fault experiment for series resistance testing
rin

262 Fig 11 is the framework diagram of a testing bench for testing battery insulation faults in a series
263 resistance test of a battery system. Connect a resistor R (replaceable or variable resistor) in series between
264 the inside of the main positive relay of the battery system and the ground. In order to calculate the
265 insulation detection time, a dual channel switch 4 is designed on circuit 1 and circuit 2. When switch 4
ep

266 is pressed, circuit 1 conducts and circuit 2 opens. BMS can confirm the starting time of insulation
267 resistance connection by detecting the on/off of circuit 2.
268 By comparing the time difference between HVIL disconnection and reporting insulation faults, the
269 response time for insulation detection can be determined. The reported insulation value can be directly
Pr

270 read out, and the experimental results are shown in Fig 12.
271 Fig 12 shows the test results of grounding the positive pole inside the battery pack and disconnecting
272 the HVIL circuit under normal vehicle operation. From the results in the figure, it can be seen that when

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 273 the HVIL circuit is disconnected, the BMS immediately reports a HVIL short circuit fault. At this time,
274 the fuzzy judgment system needs to comprehensively judge the insulation detection results by combining

ed
275 the insulation detection circuit and other sensor signals. After iterative calculation and repeated
276 verification judgment, It takes about 7.6 seconds to determine the occurrence of an insulation fault. After
277 confirming the occurrence of the insulation fault, the BMS immediately reported it. The introduction of
278 fuzzy judgment system has extended the judgment time of battery insulation detection, but significantly

iew
279 improved the accuracy of insulation detection. This experimental result shows that the introduction of
280 fuzzy fault judgment system has greatly improved the accuracy of insulation fault judgment, and
281 sacrificing a little judgment time is acceptable (the design requirement for insulation fault judgment time
282 should not exceed 10 seconds).

v
re
er
pe
ot

283
284 Fig.12 Disconnect the HVIL circuit and connect the test resistor experiment while the vehicle is in operation
285 Fig 13 shows that under normal vehicle operation, the positive pole inside the battery pack is
tn

286 grounded. As a result, the insulation resistance measured by the BMS immediately decreases from 600k
287 Ω to 0 Ω. The BMS then reports an insulation fault. When the BMS reports an insulation fault, the battery
288 system experiences high voltage under high voltage. As shown in the test results in the figure, after the
289 main positive and negative relays of the battery pack are disconnected, the fast discharge circuit of the
rin

290 motor controller connected to the external high-voltage circuit is activated, and the high voltage outside
291 the battery pack rapidly decreases. Within about 8 seconds, the voltage drops to 0V, meeting the design
292 requirements for high voltage within 10 seconds. This indicates that the designed battery fault diagnosis
293 and execution measures are effective and meet the design requirements.
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
v iew
re
er
pe
294
295 Fig.13 Experimental results of battery insulation fault during positive pole grounding test inside battery
296 pack during vehicle operation

297 6. Conclusions
ot

298 This manuscript introduces fuzzy theory into the development of fault diagnosis for power battery
299 systems, aiming to identify random battery faults that may occur under different working conditions,
300 analyze and judge whether there are real faults. While ensuring the accuracy of identification, efforts are
tn

301 made to increase the number of iterations and improve the accuracy of fault identification. Both
302 simulation and experimental data indicate that the designed fuzzy theoretical model for battery system
303 fault identification has high identification accuracy, with a slightly longer judgment time, but still meets
304 the design requirements. The research results of this article have certain reference significance for
305 research in the same industry.
rin

306
307 References
308 [1] Jiaqiang Tian，Jianning Yin. Insulation fault monitoring of lithium-ion battery pack: Recursive least
309 square with adaptive forgetting factor[J]，Journal of Energy Storage67,(2023), 107624.
ep

310 [2]. J. Tian, X. Liu, C. Chen, et al., Feature fusion-based inconsistency evaluation for battery pack:
311 improved gaussian mixture model, IEEE Trans. Intell. Transp. Syst. (2022),
312 [Link]
313 [3].Xinyang Zhang, Jichao Hong, Xiaoming Xu. Fault diagnosis of real-scenario battery systems based
Pr

314 on modified entropy algorithms in electric vehicles[J]. Journal of Energy Storage63,(2023), 107079.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 315 [4] Qiquan Liu , Jian Ma , Xuan Zhao, etal. Voltage fault diagnosis and misdiagnosis analysis of battery
316 systems using the modified Shannon entropy in real-world electric vehicles[J]. Journal of Energy Storage

ed
317 73 (2023) 109287.
318 [5] L. Jiang, Z. Deng, X. Tang, et al., Data-driven fault diagnosis and thermal runaway warning for
319 battery packs using real-world vehicle data[J], Energy 234 (2021), 121266.
320 [6] J. Hong, Z. Wang, C. Qu, et al., Investigation on overcharge-caused thermal runaway of lithium-ion

iew
321 batteries in real-world electric vehicles[J], Appl. Energy 321 (5) (2022), 119229.
322 [7] J. Hong, Z. Wang, C. Wen, et al., Online joint-prediction of multi-forward-step battery SOC using
323 LSTM neural networks and multiple linear regression for real-world electric vehicles[J], Energy Storage
324 30 (2020), 101459.
325 [8] D. Ren, X. Feng, L. Liu, et al., Investigating the relationship between internal-short circuit and

v
326 thermal runaway of lithium-ion batteries under thermal-abuse condition[J], Energy Storage Mater. 34
327 (2021) 563e73.

re
328 [9] X. Feng, Y. Pan, X. He, et al., Detecting the internal short circuit in large-format lithium-ion battery
329 using model-based fault-diagnosis algorithm[J], Energy Storage 18 (2018) 26–39.
330 [10] J. Guerrero, G. Navarro, C. Platero, et al., A novel ground fault detection method for electric vehicle
331 powertrains based on a grounding resistor voltage analysis[J], IEEE Trans. Ind. Appl. 56 (5) (2020)
332 4934–4944.
333
er
[11] X. Rui, X. Feng, H. Wang, et al., Synergistic effect of insulation and liquid cooling on mitigating
334 the thermal runaway propagation in lithium-ion battery module[J], Appl. Therm. Eng. 199 (2021),
335 117521.
pe
336 [12] H. Potdevin, Insulation monitoring in high voltage systems for hybrid and electric vehicles[J], ATZ
337 elektronik worldwide 4 (6) (2009) 28–31.
338 [13] H. Peng, Insulation and monitoring system for pure electric vehicle based on microcontroller unit[J],
339 Int. Sci. J. Acta Technica 62 (2C) (2017) 23–34.
340 [14] R. Yang, Y. Sun, H. Wu, et al., Insulation resistance measurement of voltage leakage system for EV
ot

341 battery pack[J], Int. J. Performability Eng. 16 (4) (2020) 560–568.

342 [15] J. Guerrero, G. Navarro, C. Platero, et al., A novel ground fault detection method for electric vehicle
343 powertrains based on a grounding resistor voltage analysis[J], IEEE Trans. Ind. Appl. 56 (5) (2020)
tn

344 4934–4944.
345 [16] Y. Shen, A. Liu, G. Cui, Design of online detection system for insulation resistance of electric
346 vehicle based on unbalanced bridge[C], in: 2019 IEEE Innovative Smart Grid Technologies-Asia (ISGT
347 Asia), IEEE, 2019, pp. 537–542.
rin

348 [17] K. Movassagh, A. Raihan, B. Balasingam, K. Pattipati, A critical look at the coulomb counting
349 approach for state of charge estimation in batteries, Energies 14 (14) (2021) 4074.
350 [18] Hannan MA, Lipu MSH, Hussain A, Mohamed A. A review of lithium-ion battery state of charge
351 estimation and management system in electric vehicle applications: Challenges and recommendations[J].
ep

352 Renew Sust Energ Rev 2017;78:834–54.

353 [19] Kang YZ, Duan B, Zhou ZK, Shang YL, Zhang CH. A multi-fault diagnostic method based on an
354 interleaved voltage measurement topology for series connected battery packs. [J] Power Sources
355 2019;417:132–44.
Pr

356 [20] Finegana DP, Darst J, Walker W, Li QB, Yang CB, Jervis R, et al. Modelling and experiments to
357 identify high-risk failure scenarios for testing the safety of lithium-ion cells. [J] Power Sources
358 2019;417:29–41.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 1 Design and Development of a Power Battery Insulation Fault Diagnosis System
2 Based on Fuzzy Theory

ed
3
4 Abstract: In depth exploration and research have been conducted on the identification and judgment of
5 power battery system faults. A power battery fault diagnosis and service system based on fuzzy theory
6 has been constructed. By analyzing the classification of power battery faults and the collected fault signal

iew
7 framework samples, fuzzy theory is used to analyze and identify battery system faults. Through
8 simulation and experimental comparison, it can be concluded that, The fuzzy recognition system for
9 power battery faults constructed has good recognition accuracy and short recognition time, and has good
10 engineering promotion value.
11 Keyword: Battery fault diagnosis system, Insulation, Fuzzy theory, Fault identification.

v
12
Abbreviations
13

re
14
EV electric vehicle DEM Diagnostic event manager
15
DTC Diagnostic Trouble Code BSM Basic software
16
DID Date Identifier CDD CANdela Diagnostic Descriptions
17
ODX Open Diagnostic Data Exchange SWC Software component
18
19
ECUM
DCM
ECU state manager
Diagnostic communication manager
er FIM Function inhibition manager
NVM NVRAM manager
20
RTE Run-Time Environment EDV Emergency vehicle driver
21
pe
ERA Emergency Response Action IUD Integration unit driver
22
PCM phase change material PTC positive temperature coefficient
23
EDS electric drive system BMS battery management system
24
PRA power relay assembly IPU integrated power unit
25
AC air conditioning VCU vehicle control unit
26
ot

IDI Integrated Discrimination Improvement

27

28 1. Introduction
tn

29 Power batteries are the core components of electric vehicles, and battery management
30 systems(BMS，Battery Management System) and application technologies are the diagnostic and
31 control technologies of power batteries. One of the core contents of power battery system development
32 is the ability to timely and accurately detect, identify, and judge battery faults, which can effectively
33 extend the service life of power batteries, improve their consistency and reliability. However, during
rin

34 the use of power batteries, there is complexity and ambiguity in the causes of faults, and misjudgment
35 should not affect the normal use of power batteries. [1] Therefore, it is necessary to achieve optimal
36 judgment in power battery fault diagnosis and prevention of misjudgment to ensure the safety and
37 convenience of electric vehicle use.[2]
ep

38 At present, there are two diagnostic communication protocols used by various vehicle
39 manufacturers around the world: KWP2000 and UDS (Unified Diagnostic Services). We have chosen
40 the more popular UDS diagnostic communication protocol among them. The UDS diagnostic
41 communication protocol consists of a data link layer, a network layer, and an application layer. The data
Pr

42 link layer is mainly responsible for implementing underlying CAN (Controller Area Network) driver
43 functions, the network layer implements multi frame transmission protocols, and the application layer
44 implements specific diagnostic services and corresponding communication protocols. The fault handling

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 45 function for specific faults can be roughly divided into several parts, including fault detection, fault
46 management, and system fault response and handling. [3-4]

ed
47 The application environment of power batteries in electric vehicles is very complex, and there are
48 various interference factors. Therefore, it is very important to timely and accurately judge the fault
49 information. Only by accurately judging various fault information can have targeted and effective
50 measures be taken to reduce fault injuries. To solve this problem, it is necessary to construct a diagnostic

iew
51 model for battery faults based on existing battery big data fault information, combined with high-
52 precision mathematical models obtained from previous experiments. By combining various sensor
53 signals and interference information in the vehicle, key judgment factors can be correctly identified, fault
54 information can be located, and effective processing measures can be taken to achieve fault isolation.

55 2. BMS Development of diagnostic functions

v
56 The main goal of developing BMS application software is to monitor battery information and ensure
57 the normal use of the battery system; At the same time, on the basis of strengthening management, strive

re
58 to improve battery efficiency as much as possible and ensure safe use. [5-6]
59 The UDS diagnostic protocol is a requirement specification for diagnostic communication in the
60 automotive industry, defined by the ISO-14229 series of standards. The application layer (Layer 7)
61 applied to the OSI seven-layer model only specifies service requirements related to diagnosis and does
er
62 not involve communication mechanisms. Therefore, it can be implemented on different automotive buses
63 (such as CAN, LIN, Flexray, Ethernet, and K-line). [7-8]
64 Diagnostic service is an information exchange method between diagnostic equipment and ECU
pe
65 (Electronic Control Unit) [9] Usually, a diagnostic device sends a request and the ECU responds. A fault
66 code is a code used to mark ECU faults, which follows certain rules and is stored in the non-volatile
67 storage of the ECU. Diagnostic data is the internal data of the ECU that can be requested by diagnostic
68 equipment, including: current data: data that the ECU is running, such as vehicle speed, throttle opening,
69 engine speed, etc; Storage data: The data stored in the memory by the ECU at a certain time, such as
ot

70 DTC; Static data: Constant and unchanging ECU internal data, such as VIN codes. [10]
tn
rin
ep

71
72 Fig.1 Definition of Application Layer Timing Parameters for UDS Diagnostic Protocol
Pr

73 The implementation phase of BMS fault diagnosis software includes implementing fault detection
74 function in TargetLink model, fault response and processing function in TargetLink model, DTC
75 (Diagnostic Trouble Code) storage function in C code, fault related diagnostic service function in C code,

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 76 and software integration from UDS protocol stack code to application layer C code and TargetLink model
77 in C code.[11] The specific process is showed in Fig 2.

ed
Implementing Implement
Implementing Implement software
fault response Implement fault related
fault detection integration from UDS
and handling DTC storage diagnostic
function in the protocol stack code to
functions in the function in C service
TargetLink application layer C code
TargetLink code functions in
model and TargetLink model in C

iew
model C code
78 code
79 Fig.2 BMS software fault diagnosis design and development process
80 The testing and verification phase is to verify whether the BMS system fault diagnosis strategy
81 software implementation meets the UDS protocol specifications and fault diagnosis strategy
82 specifications, mainly including UDS protocol CAN delta automatic testing, fault DTC detection and

v
83 storage HIL testing, fault response HIL testing, joint testing with on-board diagnostic equipment, and
84 joint testing with offline testing equipment.

re
UDS
Diagnostic Joint testing
Fault DTC Fault
Protocol of onboard Offline detector joint
detection and handling HIL
CANdela diagnostic testing
storage HIL test testing
automatic equipment
testing er
85
86 Fig.3 Diagnostic service testing process

87 3. A Fuzzy System Based Algorithm Model for Battery Fault Diagnosis

pe
88 The electrochemical model of lithium-ion battery is an electrochemical equation model that can
89 characterize the characteristics of lithium-ion battery through comprehensive analysis of the internal
90 working mechanism of lithium-ion battery, the material of lithium-ion battery, the spectrum of lithium-
91 ion battery and the internal diffusion effect. [12-15]

92 {𝑈𝑑𝑈= 𝑈𝑑𝑡 =(𝑆𝑂𝐶)

𝐿 ― 𝑖(𝑡)𝑅 ― 𝑈
𝑂𝐶
―𝑈 𝑅 𝐶 +𝑖 𝐶
𝑂 𝑃
(1)
ot

𝑃 𝑃 𝑃 𝑃 𝑃

93 𝑈𝑃𝑏 = ∫ 𝑖𝑑𝑡 + 𝑈𝑃𝑏0 (2)

94 In the formula, UL is the terminal voltage, Uoc is the open-circuit voltage, i (t) is the load current,
tn

95 R0 is the ohmic internal resistance, Cp is the equivalent capacitance of the electrode plate, Up is the
96 terminal voltage at both ends of the polarization capacitor Cp, i is the current, Cb is the equivalent
97 capacitance describing the open-circuit voltage change, and Upb is the terminal voltage at both ends of
98 the capacitor Cb.
rin

⌈ ⌉
∆𝑡

{
―𝜂

99
⌈ 𝑆𝑂𝐶𝑘
𝑈𝑝,𝑘 ⌉=[ 1
0
𝑒
0
―∆𝑡
𝜏
][ 𝑆𝑂𝐶𝑘―1
𝑈𝑝,𝑘―1 ]+ 𝑅𝑝(1 ― 𝑒
𝐶𝑚𝑎𝑥
―∆𝑡
𝜏 )
𝑖𝑘 + 𝑤
(3)
𝑈𝐿,𝑘 = 𝑈𝑜𝑐(𝑆𝑂𝐶𝑘) ― 𝑈𝑝,𝑘 ― 𝑖𝑘𝑅𝑜 + 𝜈
ep

100 In the formula, Cmax is the maximum usable capacity of the battery, η is the coulomb efficiency,
101 ∆ t is the sampling period, τ = 𝑅𝑃𝐶𝑃is the time constant, w is the process noise, and v is the
102 observation noise.
𝜂𝑖Δ𝑡𝑖(𝑘)
Pr

103 SOC(𝑘 + 1) = SOC(𝑘) ― (1 ― 𝜂𝑛)𝐶𝑛 (4)

𝜂𝑇
104 In the formula, SOC (k+1) is the SOC value of the battery at the time of k+1; I (t) is the current
105 value at the time of k~k+1; Qn is the rated capacity of the battery; η i， η T and η N represents the

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 106 correction coefficient of the charge-discharge ratio, temperature and charge-discharge cycle number on
107 the battery capacity, which can be obtained by processing the data of the charge-discharge experiment

ed
108 of the battery.[16]
1

{ [] ( [ ]
0 0

109
𝑈𝑏
𝑈𝑝
= 0 ―1 𝐶𝑝𝑅𝑝 )[ 𝑈𝑏
𝑈𝑝]+ 1
𝐶𝑏
𝐶𝑝
𝐼𝐿
(5)

iew
[𝑈𝐿] = [ ―1 [ ]
𝑈
―1] 𝑈𝑏 + [ ― 𝑅𝑂][𝐼𝐿] + [𝑈𝑂𝐶]
𝑝

110 In the model, the voltage Ub and Up is the state variable of the two capacitors, and the terminal
111 voltage UL of the battery is the output variable.
112 Fuzzy neural networks are neural networks with fuzzy weight coefficients or input signals that are

v
113 fuzzy quantities. Fuzzy neural networks, whether used as approximators or pattern memory, require
114 learning and optimizing weight coefficients. Learning algorithms are the key to optimizing weight

re
115 coefficients in fuzzy neural networks. For logical fuzzy neural networks, error based learning algorithms,
116 also known as supervised learning algorithms, can be used. For arithmetic fuzzy neural networks, there
117 are fuzzy BP algorithm, genetic algorithm, etc. There is currently no reasonable algorithm for hybrid
118 fuzzy neural networks; However, hybrid fuzzy neural networks are generally used for computation rather
119 than learning, and they do not necessarily need to be learned. [17]
120
er
The most effective and widespread application of fuzzy theory is in fuzzy control, which
121 unexpectedly solves problems that traditional control theory cannot solve or is difficult to solve in various
122 fields, and has achieved some convincing results. Summarize the control strategies of human experts for
pe
123 specific controlled objects or processes into a series of control rules expressed in the form of "IF
124 (condition) THEN (action)", obtain the control action set through fuzzy reasoning, and apply it to the
125 controlled object or process. The control action set is a set of conditional statements, while the state
126 statement and control action are both quantified fuzzy language sets. [18]
ot
tn
rin

127
128 Fig.4 The relationship between the number of iterations, Spread membership, crossover probability, and
ep

129 mutation probability membership functions of fuzzy functions

Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
iew
130

v
131 Fig.5 Fuzzy System Output Diagram of Cross Probability

re
132 The diversity, convergence, and number of iterations of the sample population are closely related
133 to the performance indicators of the algorithm and the algorithm results. Therefore, the number of
134 iterations and convergence indicators are used as inputs to the fuzzy system, and the crossover probability
135 and mutation probability are dynamically adjusted during each iteration process. In order to measure the
136 number of iterations of the algorithm, the iteration value is limited to the interval [0,1], and the
137
er
convergence metric spread is normalized. The input of each sample is granulated into three triangular
138 membership functions (low, mid, and high respectively represent the size of the input value), and the
139 output is granulated into five triangular membership functions (low, mid low, mid, mid high, high
pe
140 respectively represent the size of the output value)。
ot
tn
rin
ep

141
142 Fig.6 Fault handling flowchart based on Fuzzy System
Pr

143 Fig 6 is a fault diagnosis and judgment flowchart based on the application of fuzzy theory. The fault
144 diagnosis of the vehicle controller mainly includes fault detection, fault management, and fault handling.
145 The specific diagnostic process is shown in Figure 6. Firstly, based on the VCU input/output interface

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 146 and functional definition, fault detection is carried out to confirm the specific faults currently present in
147 the VCU; Then, for confirmed faults, they will be automatically stored at the bottom layer based on pre-

ed
148 defined fault codes and frozen frames; Finally, different faults will directly correspond to corresponding
149 fault handling actions, and each functional module will then perform specific failure protection actions.
150 As shown in Fig6, DFC file is diagnostic fault check file while FID is Fault Identification
151 Denotation. After receiving various fault information, the processor will make judgments and analyze

iew
152 processing in the fault detection module, form corresponding menu files, and predict the next step of
153 fault management information by comparing the corresponding fault code information based on the FRM
154 model. In the fault management information module, based on the judgment logic and calibration
155 parameters constructed from the corresponding DTC and DID lists, corresponding fault levels and
156 expected fault actions are divided, In the fault handling module, the driver's behavior and vehicle status

v
157 will be comprehensively judged based on the obtained fault management information and corresponding
158 fault levels, and corresponding fault actions will be executed based on various types of information. The

re
159 entire fault diagnosis and handling process is constantly being repeated and confirmed based on changes
160 in driving conditions.

161 4. Identification and isolation of Battery Faults

162 The signals and communication information collected from electric vehicles are mixed with a large
er
163 amount of error noise. In order to reduce the probability of false alarms and improve detection accuracy,
164 it is necessary to set some judgment rules to determine whether corresponding faults have actually
165 occurred. In this study, the collected CAN communication signals and sensor fault information are used
pe
166 as input samples for the fuzzy system, and corresponding estimation thresholds are formulated, To
167 analyze and judge the corresponding fault information, and finally confirm whether the fault has actually
168 occurred. Only after confirmation can the corresponding fault handling measures be implemented. [19]
169 After BMS detects a fault, it will start an appropriate debounce process. Both fault confirmation
170 and fault recovery must go through a calibrated debounce time. The Debounce module is a timer that
ot

171 starts running when the enable signal MonEna is 1, and the program can control its reset. After reset or
172 initialization, the timer value will be set to 0. [20]
173 If the monitoring result MonDfct of the fault detection report is 1, it indicates that a fault has
tn

174 occurred and the "fail" timer increases. Once this timer reaches the calibratable threshold of "fail" (tiDfct),
175 bit 0, bit 1, and bit 3 of the status word stDbc will be set to 1.
176 If the monitoring result MonDfct of the fault detection report is 0, that is, no fault has occurred, the
177 "pass" timer will increase. Once this timer reaches the calibrated threshold (tiOk) for "pass", the bit 0 of
178 the status word stDbc will be reset to 0.
rin

179 The anti shake debounce module of HCU follows the following rules: the debounce module is
180 designed based on a counter, which increases when the reported fault monitoring result MonOk is 1. If
181 the reported fault monitoring result is 0, it indicates that no fault has been detected, and this counter
182 decreases.
ep

183 Due to this software structure, fault detection "pass" and "fail" can never occur simultaneously.
184 Tab 1. Main battery failure threshold table
Fault Clear vehicle operation Vehicle standing still
Diagnostic Fault
Pr

Fault Fault toler Diagnos fault Recovery External

Item start recovery Dsichar Dsichar
name threshold ance tic cycle conditio time Charge charging Charg
conditions value ge ge
time ns e

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 >4.20V 4s 10ms ≤4.18V 2s Map 0 Stop Stop enable

ed
cel
>4.25V 4s 10ms ≤4.2V 2s 10kW 0 cut off Stop enable

voltage
unable
From
too high
>4.30V 1s 10ms to / Power down cut off Stop Stop
initializatio
Automat
recover
cell n to low

iew
ic
≤2.8V 4s 10ms >2.9V 2s 10kW Map enable enable enable
voltage voltage
clearing
Request to power
power
cell ≤2.5V 2s 10ms >2.8V 2s enable enable Stop
down
down
voltage
unable
too low
≤2.0V 1s 10ms to / Power down Stop Stop Stop

recover

v
Module ≥50℃ 4s 10ms ≤45℃ 4s Map Map enable enable enable

re
temperatu Request to power
≥55℃ 4s 10ms ≤50℃ 4s Stop Stop Stop
re too down
Module
high ≥60℃ 4s 10ms ≤55℃ 4s Power down Stop Stop Stop
temperatu / /
≤-20℃ 4s 10ms ≥-17℃ 4s Map Map Stop Stop enable
re Module
≤-25℃ 4s 10ms ≥-22℃ 4s Map Map Stop Stop enable
temperatu

re too low ≤-30℃ 4s

er
10ms ≥-27℃ 4s
Request to power
Stop Stop Stop
down

<0.95*
pe
From
>0.9*DCH DCH immediatel
discharge 2s initializatio 10ms 10kW Map / enable enable
threshold threshol y
current n to low
d
overcurre voltage
<DCH
nt ≥DCH power Request to power
2s 10ms / threshol 100ms / enable enable
ot

threshold down down

d

Short

circuit of 500 Request to power

tn

≥550A 10ms <550A 100ms / enable enable

discharge ms down
From
current
current initializatio
<0.95*R
n to high
>0.9*RCH CH immediatel Request to power
rin

2s voltage 10ms / enable enable

Recharge threshold threshol y down
power
overcurre d
down
nt <RCH
≥RCH Request to power
2s 10ms / threshol 100ms / enable enable
ep

threshold down
d

From <0.95*C
Charging
>0.9*CH initializatio H immediatel
overcurre 2s 10ms / / / enable enable
threshold n to stop threshol y
Pr

nt
charging d

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 <CH
≥CH
2s 10ms threshol 100ms / / / /

ed
threshold
d

≥1250Ω
≤1000Ω/V 20s 10ms 20s / / enable enable enable
/V

AC ≥1000Ω

iew
≤750Ω/V 20s 10ms 20s / / enable enable enable
insulation From /V

initializatio ≥550Ω/
≤300Ω/V 20s 10ms 20s / / Stop Stop Stop
n to high V

voltage ≥1250Ω
≤1000Ω/V 20s 10ms 20s Map Map / enable enable
DC power /V

v
insulation down ≥1000Ω
insulation ≤750Ω/V 20s 10ms / 20s 10kW 10kW / enable enable
(closed /V

re
relay) ≥750Ω/ Request to power
≤500Ω/V 20s 10ms 20s / Stop Stop
V down

From ≥1250Ω
≤1000Ω/V 20s 10ms 20s Map Map / enable enable
DC initializatio /V

insulation

(open
≤750Ω/V 20s
n to low

voltage
er
10ms
≥1000Ω

/V
20s 10kW 10kW / enable enable

relay) power ≥750Ω/ Request to power

≤500Ω/V 20s 10ms 20s / Stop Stop
pe
down V down

185 Table 1 is the compiled threshold table for power battery faults, which is used to determine whether
186 a fault has occurred by combining the control thresholds in Table 1 with relevant auxiliary signals. Taking
187 the insulation fault diagnosis in the table as an example, after detecting a signal with a low insulation
188 resistance value, due to the CAN bus communication cycle of the insulation fault being 1 second, the
189 BMS needs to receive a signal with a low insulation resistance value for 20 consecutive times, add other
ot

190 auxiliary judgment information, and comprehensively determine the insulation fault before executing the
191 corresponding reporting and processing measures. For example, if the insulation resistance value is ≤
192 1000 Ω/V, the BMS will report a level one minor insulation fault information, and the vehicle can still
tn

193 be used normally. The charging and discharging power map remains unchanged. When the insulation
194 resistance value is ≥ 1250 Ω/V, the fault will automatically recover; When the insulation resistance
195 value is ≤ 750 Ω /V, the BMS will report a secondary insulation fault information. The battery
196 discharge power and feedback power are both limited to 10kW, and the whole vehicle enters a limp state.
rin

197 At this time, the charging and discharging power of the battery is very small, and the discharge power is
198 only enough for moving the vehicle. When the insulation resistance value is ≥ 1000 Ω/V, the fault
199 will automatically recover; When the insulation resistance value is ≤ 500 Ω/V, the BMS will report
200 a three-level insulation fault information. The battery discharge power and feedback power are limited
ep

201 and cannot be used, and the entire vehicle is prohibited from running. The BMS requests to power off.
202 When the insulation resistance value is ≥ 750 Ω/V, the fault will automatically recover.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
v iew
re
203
204 Fig.7 Insulation resistance simulation circuit diagram ( DC side)
205 Fig 7 is an insulation simulation model diagram of the high-voltage battery part in the power battery
206 high-voltage circuit system established in Simulink software. In this simulation model, the BMS is first
207
er
awakened to perform the power on operation. After the BMS is powered on, at 3 seconds, a connection
208 resistance of 100k Ω is introduced to determine whether the entire insulation detection system can detect
209 insulation faults inside the battery pack in a timely manner, And whether the fault handling logic executed
pe
210 is consistent with the corresponding fault control threshold, and whether the processing flow of the
211 actuator conforms to the designed fault handling flow. Does the time for fault diagnosis and response
212 time meet the design requirement?
213 Fig 8 shows the results obtained from simulation calculations. From the results in Figure 8, it can
214 be seen that after the BMS wakes up and starts normally, the self check passes, and the BMS state value
ot

215 is referred to as 2 (main relay closed). The BMS enters the working state, and after calibration at the 3
216 second position, a resistor with a series resistance value of 100k Ω is introduced. After about 3 seconds
217 of judgment time, the BMS detects an insulation fault and reports the insulation fault status. In a relatively
tn

218 short time of about 150ms, BMS reports a fault, and the status value of the BMS is set to 9 (power hold
219 relay open) in a short period of time. After repeated confirmation for about 100ms, the BMS reports an
220 insulation fault, and the entire simulation result meets the design requirements of the control threshold
221 table. The response time meets the design requirements. Therefore, the simulation results prove that after
rin

222 introducing relevant series resistors, the BMS can make accurate fault diagnosis results and execute
223 actions.
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
iew
224

v
225 Fig.8 Simulation result of battery Insulation resistance based on Fuzzy System

226 5. Discussion on Experiment results

re
er
pe
ot

227
tn

228 Fig.9 BMS fault test and insulation fault test platform
229 Fig 9 is an experimental setup diagram of the experimental setup. Build the corresponding test
230 bench according to the testing requirements. The upper left image in the figure shows the HIL detection
231 equipment, which simulates various faults and communication signals of the power battery system
rin

232 through the HIL experimental equipment to verify the recognition rate of the constructed fuzzy judgment
233 system. The number of test samples in this experiment is set to 100, which is also close to the number of
234 fault samples tested in actual vehicles. Therefore, the results of this setting are more representative. The
235 upper middle position of Figure 6 is an experimental device that simulates various input and
ep

236 communication signal errors and interference signals of the BMS. The accuracy of the BMS model
237 judgment is analyzed and judged through the simulated signals. The lower left corner of Figure 6 shows
238 the testing bench where the actual measurement signal error of the battery pack is introduced. By
239 introducing measurement errors from various sensors, the accuracy of fault diagnosis in the simulation
Pr

240 system is analyzed and tested. The right part of Figure 9 is a device for measuring insulation faults in
241 real vehicles, which determines the accuracy of insulation faults by introducing insulation values of
242 various resistance values.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 243 Fig 10 shows the recognition and calculation results of insulation faults and overcurrent faults in
244 the fault signal of the constructed fuzzy system. In the first 50 seconds, no faults are input. After 50

ed
245 seconds, after introducing fault signals and interference signals, and separately introducing insulation
246 fault signals and overcurrent fault signals, it can be seen from Figure 7 that in the first 50 seconds without
247 faults, the fault model values and the system's judgment values are very consistent, without residual
248 values; After introducing simulated fault signals and interference signals, there was a deviation between

iew
249 the system's judgment value and input fault value, forming a certain residual value. However, the residual
250 values were very small, both within 1.0. After about 220 seconds, the insulation fault residual value
251 approached 0, and the model error and judgment value error basically disappeared; After about 330
252 seconds, the overcurrent error value and judgment value error basically disappear. After this, the fuzzy
253 recognition system can basically recognize the corresponding fault signal content. After repeated testing,

v
254 the accuracy of the model's judgment error is about 92.7%, which is an ideal result.

re
er
255
pe
256 Fig.10 Residual of insulation fault model based on Fuzzy System
ot
tn

257
258 Fig.11 Schematic diagram of insulation fault experiment for series resistance testing
rin

259 Fig 11 is the framework diagram of a testing bench for testing battery insulation faults in a series
260 resistance test of a battery system. Connect a resistor R (replaceable or variable resistor) in series between
261 the inside of the main positive relay of the battery system and the ground. In order to calculate the
262 insulation detection time, a dual channel switch 4 is designed on circuit 1 and circuit 2. When switch 4
ep

263 is pressed, circuit 1 conducts and circuit 2 opens. BMS can confirm the starting time of insulation
264 resistance connection by detecting the on/off of circuit 2.
265 By comparing the time difference between HVIL disconnection and reporting insulation faults, the
266 response time for insulation detection can be determined. The reported insulation value can be directly
Pr

267 read out, and the experimental results are shown in Fig 12.
268 Fig 12 shows the test results of grounding the positive pole inside the battery pack and disconnecting
269 the HVIL circuit under normal vehicle operation. From the results in the figure, it can be seen that when

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 270 the HVIL circuit is disconnected, the BMS immediately reports a HVIL short circuit fault. At this time,
271 the fuzzy judgment system needs to comprehensively judge the insulation detection results by combining

ed
272 the insulation detection circuit and other sensor signals. After iterative calculation and repeated
273 verification judgment, It takes about 7.6 seconds to determine the occurrence of an insulation fault. After
274 confirming the occurrence of the insulation fault, the BMS immediately reported it. The introduction of
275 fuzzy judgment system has extended the judgment time of battery insulation detection, but significantly

iew
276 improved the accuracy of insulation detection. This experimental result shows that the introduction of
277 fuzzy fault judgment system has greatly improved the accuracy of insulation fault judgment, and
278 sacrificing a little judgment time is acceptable (the design requirement for insulation fault judgment time
279 should not exceed 10 seconds).

v
re
er
pe
ot

280
281 Fig.12 Disconnect the HVIL circuit and connect the test resistor experiment while the vehicle is in operation
282 Fig 13 shows that under normal vehicle operation, the positive pole inside the battery pack is
tn

283 grounded. As a result, the insulation resistance measured by the BMS immediately decreases from 600k
284 Ω to 0 Ω. The BMS then reports an insulation fault. When the BMS reports an insulation fault, the battery
285 system experiences high voltage under high voltage. As shown in the test results in the figure, after the
286 main positive and negative relays of the battery pack are disconnected, the fast discharge circuit of the
rin

287 motor controller connected to the external high-voltage circuit is activated, and the high voltage outside
288 the battery pack rapidly decreases. Within about 8 seconds, the voltage drops to 0V, meeting the design
289 requirements for high voltage within 10 seconds. This indicates that the designed battery fault diagnosis
290 and execution measures are effective and meet the design requirements.
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 ed
v iew
re
er
pe
291
292 Fig.13 Experimental results of battery insulation fault during positive pole grounding test inside battery
293 pack during vehicle operation

294 6. Conclusions
ot

295 This manuscript introduces fuzzy theory into the development of fault diagnosis for power battery
296 systems, aiming to identify random battery faults that may occur under different working conditions,
297 analyze and judge whether there are real faults. While ensuring the accuracy of identification, efforts are
tn

298 made to increase the number of iterations and improve the accuracy of fault identification. Both
299 simulation and experimental data indicate that the designed fuzzy theoretical model for battery system
300 fault identification has high identification accuracy, with a slightly longer judgment time, but still meets
301 the design requirements. The research results of this article have certain reference significance for
302 research in the same industry.
rin

303
304 References
305 [1] Jiaqiang Tian，Jianning Yin. Insulation fault monitoring of lithium-ion battery pack: Recursive least
306 square with adaptive forgetting factor[J]，Journal of Energy Storage67,(2023), 107624.
ep

307 [2]. J. Tian, X. Liu, C. Chen, et al., Feature fusion-based inconsistency evaluation for battery pack:
308 improved gaussian mixture model, IEEE Trans. Intell. Transp. Syst. (2022),
309 [Link]
310 [3].Xinyang Zhang, Jichao Hong, Xiaoming Xu. Fault diagnosis of real-scenario battery systems based
Pr

311 on modified entropy algorithms in electric vehicles[J]. Journal of Energy Storage63,(2023), 107079.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]
 312 [4] Qiquan Liu , Jian Ma , Xuan Zhao, etal. Voltage fault diagnosis and misdiagnosis analysis of battery
313 systems using the modified Shannon entropy in real-world electric vehicles[J]. Journal of Energy Storage

ed
314 73 (2023) 109287.
315 [5] L. Jiang, Z. Deng, X. Tang, et al., Data-driven fault diagnosis and thermal runaway warning for
316 battery packs using real-world vehicle data[J], Energy 234 (2021), 121266.
317 [6] J. Hong, Z. Wang, C. Qu, et al., Investigation on overcharge-caused thermal runaway of lithium-ion

iew
318 batteries in real-world electric vehicles[J], Appl. Energy 321 (5) (2022), 119229.
319 [7] J. Hong, Z. Wang, C. Wen, et al., Online joint-prediction of multi-forward-step battery SOC using
320 LSTM neural networks and multiple linear regression for real-world electric vehicles[J], Energy Storage
321 30 (2020), 101459.
322 [8] D. Ren, X. Feng, L. Liu, et al., Investigating the relationship between internal-short circuit and

v
323 thermal runaway of lithium-ion batteries under thermal-abuse condition[J], Energy Storage Mater. 34
324 (2021) 563e73.

re
325 [9] X. Feng, Y. Pan, X. He, et al., Detecting the internal short circuit in large-format lithium-ion battery
326 using model-based fault-diagnosis algorithm[J], Energy Storage 18 (2018) 26–39.
327 [10] J. Guerrero, G. Navarro, C. Platero, et al., A novel ground fault detection method for electric vehicle
328 powertrains based on a grounding resistor voltage analysis[J], IEEE Trans. Ind. Appl. 56 (5) (2020)
329 4934–4944.
330
er
[11] X. Rui, X. Feng, H. Wang, et al., Synergistic effect of insulation and liquid cooling on mitigating
331 the thermal runaway propagation in lithium-ion battery module[J], Appl. Therm. Eng. 199 (2021),
332 117521.
pe
333 [12] H. Potdevin, Insulation monitoring in high voltage systems for hybrid and electric vehicles[J], ATZ
334 elektronik worldwide 4 (6) (2009) 28–31.
335 [13] H. Peng, Insulation and monitoring system for pure electric vehicle based on microcontroller unit[J],
336 Int. Sci. J. Acta Technica 62 (2C) (2017) 23–34.
337 [14] R. Yang, Y. Sun, H. Wu, et al., Insulation resistance measurement of voltage leakage system for EV
ot

338 battery pack[J], Int. J. Performability Eng. 16 (4) (2020) 560–568.

339 [15] J. Guerrero, G. Navarro, C. Platero, et al., A novel ground fault detection method for electric vehicle
340 powertrains based on a grounding resistor voltage analysis[J], IEEE Trans. Ind. Appl. 56 (5) (2020)
tn

341 4934–4944.
342 [16] Y. Shen, A. Liu, G. Cui, Design of online detection system for insulation resistance of electric
343 vehicle based on unbalanced bridge[C], in: 2019 IEEE Innovative Smart Grid Technologies-Asia (ISGT
344 Asia), IEEE, 2019, pp. 537–542.
rin

345 [17] K. Movassagh, A. Raihan, B. Balasingam, K. Pattipati, A critical look at the coulomb counting
346 approach for state of charge estimation in batteries, Energies 14 (14) (2021) 4074.
347 [18] Hannan MA, Lipu MSH, Hussain A, Mohamed A. A review of lithium-ion battery state of charge
348 estimation and management system in electric vehicle applications: Challenges and recommendations[J].
ep

349 Renew Sust Energ Rev 2017;78:834–54.

350 [19] Kang YZ, Duan B, Zhou ZK, Shang YL, Zhang CH. A multi-fault diagnostic method based on an
351 interleaved voltage measurement topology for series connected battery packs. [J] Power Sources
352 2019;417:132–44.
Pr

353 [20] Finegana DP, Darst J, Walker W, Li QB, Yang CB, Jervis R, et al. Modelling and experiments to
354 identify high-risk failure scenarios for testing the safety of lithium-ion cells. [J] Power Sources
355 2019;417:29–41.

This preprint research paper has not been peer reviewed. Electronic copy available at: [Link]