Article

Insulation Monitoring Method for DC Systems with

Ground Capacitance in Electric Vehicles
Jifei Du 1, Trillion Q. Zheng 1, Yian Yan 2,*, Hongyan Zhao 1, Yangbin Zeng 1 and Hong Li 1
1 School of Electrical Engineering, Beijing Jiaotong University, No. 3 Shang Yuan Cun, Haidian District,
Beijing 100044, China; 13117363@bjtu.edu.cn (J.D.); tqzheng@bjtu.edu.cn (T.Q.Z.);
hyzhao2013@126.com (H.Z.); ybzeng@bjtu.edu.cn (Y.Z.); hli@bjtu.edu.cn (H.L.)
2 National Active Distribution Network Technology Research Center (NANTEC),

Beijing Jiaotong University, Beijing 100044, China; yanyian@bjtu.edu.cn

* Correspondence: yanyian@bjtu.edu.cn; Tel.: +86-10-5168-4056; Fax: +86-10-5168-3907

Received: 21 May 2019; Accepted: 25 June 2019; Published: 27 June 2019

Featured Application: The insulation monitoring generally applied to DC charging piles,

battery management systems, or high voltage distribution system for electric vehicles. It is
effective on improving the safety performance of electric vehicles.

Abstract: Owing to the influence of ground capacitance in electric vehicles, in the traditional
unbalanced electric bridge DC insulation monitoring (DC-IM) method, the voltage of positive and
negative electric bridges changes slowly. To calculate the insulation resistances, sampling should
be conducted once the voltage of the bridge becomes stable, that will inevitably extend the
monitoring cycle. To reduce the monitoring cycle, this study proposes a three-point climbing
algorithm, namely, three-bridge voltage sampling with equal sampling intervals, to predict the
evolution of the bridge voltage curve. However, due to the existence of sampling errors, the
insulation resistances calculated by sampling values will deviate from the actual values. Then, this
article also proposes the filter and correction methods of three sampled voltages to improve
monitoring accuracy. Through experimental data, the influences of different parameters on the
results are verified, and comparisons with the traditional method are shown in the back. The
conclusion is that compared with the traditional method, the proposed method can monitor
insulation resistance more quickly and ensure fixed monitoring cycles under different ground
capacitance values and keep the similar monitoring accuracy.

Keywords: unbalanced electric bridge; DC insulation monitoring; ground capacitance; three-point

climbing algorithm

1. Introduction
With the increasing popularity of electric vehicles (EVs), strict requirements are being
established for the driving and charging safety of EVs. The insulation of EVs decreases due to rain,
dampness, collision, and other reasons arising from the long-term exposure of EVs and charging
equipment to the outdoor environment [1,2]. The DC system of EVs is connected to numerous power
electronic devices, including the motor converter, battery charger, air conditioner, and DC–DC
converter [3], and the overall connections may form a DC power microgrid system when EV is
charging [4,5]. Insulation failure of any equipment affects the safety of the entire system. When the
insulation resistance of the system decreases to below a threshold value, the vehicle sends warning
signals. If the situation is serious, then the high-voltage system must be cut off and stopped for
troubleshooting [6,7]. The DC insulation monitoring (DC-IM) function is thus required before

Appl. Sci. 2019, 9, 2607; doi:10.3390/app9132607 www.mdpi.com/journal/applsci

Appl. Sci. 2019, 9, 2607 2 of 16

charging by the DC charging pile and during the process of driving the EVs [8]. Various
insulation-monitoring devices and embedded circuits have been designed and installed in DC
charging piles, battery packs, high-voltage distribution boxes, and other equipment or embedded in
the battery management system of EVs. DC-IM methods include balanced electric bridge [9],
unbalanced electric bridge [10–14], high-voltage injection [15], differential amplification [16], and
low-frequency small-signal injection [17,18]. The unbalanced electric bridge method can
synchronously monitor positive and negative insulation resistances, has a low cost, and is easy to
realize; thus, it has been widely used in EVs and charging piles.
Because the DC system of EVs connects various power electronic devices that contain many Y
capacitors and parasitic capacitors, which make up the large ground capacitance (GC) of the system,
GC, an unknown system parameter, seriously affects the monitoring accuracy and speed of DC-IM.
Therefore, various solutions have been suggested in the literature. In [19,20], wavelet-transform and
chaos theory detection methods were proposed to deal with interference signals. However, these
methods are more suitable for a multi-branch DC system with the small-signal injection method
than for systems with a large GC. The method based on the Kalman filter and Lyapunov equation
proposed in [18] and [21] needs to be recursive step by step. Thus, obtaining the result takes a long
time. The traditional sampling and comparison method is frequently used in current practical
product applications. After initiating bridge conversion, sampling and calculation are performed
only after GC is fully charged so that a stable voltage signal can be sampled. However, this method
considerably slows down DC-IM and cannot meet the real-time requirements of EVs and the future
development trend of EV safety.
This study proposes a method of unbalanced electric bridge DC-IM based on a three-point
climbing algorithm. After switching the bridge, sampling is conducted for three times at equal
intervals. The methods of filtering and automatic correction of sampling voltage are also proposed to
reduce the result error caused by voltage ripple and sampling resolution. The calculation can predict
the voltage value after the completion of GC charging. The method is simple and easy to implement.
It does not need to wait for GC charging and multiple sampling, which can considerably increase the
detection speed. Thus, the DC-IM period is fixed and unaffected by GC.
The rest of this paper is organized as follows. Section 2 analyzes the unbalanced electric bridge
DC-IM with the existing GC. Section 3 proposes the novel method of the three-point climbing
algorithm in order to avoid the impact on GC. Section 4 further optimizes the proposed method and
describes the implementation method. In Section 5, The experimental data are exhibited to prove the
theory. Finally, conclusions are included in Section 6. Some symbols used in the operation
optimization are shown in Table 1.

Table 1. List of some symbols used in the operation optimization.

Symbol Explanation
DC-IM DC insulation monitoring
GC Ground capacitance
Rf1 The negative insulation resistance
Rf2 The positive insulation resistance
Ra The larger bridge resistance
Rb The smaller bridge resistance
C1 The capacitance value of the DC negative pole to the earth
C2 The capacitance value of the DC positive pole to the earth
v1 The voltage value of the DC negative pole to the earth
v2 The voltage value of the DC positive pole to the earth
v11 and v12 The sample value of v1 in M1 and M2 phase respectively
v21 and v22 The sample value of v2 in M1 and M2 phase respectively
Δt The time interval of sampling
Δt
-
E1 e τ1
of M1 phase
Appl. Sci. 2019, 9, 2607 3 of 16

Δt
-
E2 e τ 2 of M2 phase
Eˆ1 (k ) The estimated value of the kth E1
Eˆ (k )
2
The estimated value of the kth E2
v Estimated voltage value of the half-bridge voltage
v̂ Correction voltage value of the half-bridge voltage

2. Traditional Unbalanced Electric Bridge DC-IM Method and Analysis of the Influence on GC
The unbalanced electric bridge DC-IM topology circuit is shown in Figure 1, where vdc is the DC
voltage; Rf2 and Rf1 are the positive and negative insulation resistances, respectively; Ra and Rb are the
bridge resistances, Ra = R1, Rb = R1||R2, so Ra > Rb. The unbalanced electric bridge method works in
two phases, namely, M1 and M2. In the M1 phase, Q1 turn-on and Q2 turn-off, the positive half-bridge
resistance is Ra, the negative half-bridge resistance is Rb, the negative and positive half-bridge
voltages are v11 and v12, respectively, and the ground current is i1, as shown in Figure 1a. In the M2
phase, Q1 turn-off and Q2 turn-on, the positive half-bridge resistance is Rb, and the negative
half-bridge resistance is Ra. The negative and positive half-bridge voltages are v21 and v22,
respectively, and the ground current is i2, as shown in Figure 1b. The conventional DC-IM method
can be expressed as Equation (1) by Kirchhoff’s law.

v12 v11 v22 v21

DC+ DC- DC+ DC-

R1 R1 R1 R1

Q2 R2 R2 Q1 Q2 R2 R2 Q1

Rf2 Rf1 Rf2 Rf1

(a) (b)
vdc vdc

Ra Rb Rb Ra

i1 i2
Rf2 Rf1 Rf2 Rf1

v12 v11 v22 v21

(c) (d)
Figure 1. The circuit of unbalanced bridge. (a) The circuit of M1 phase. (b) The circuit of M2 phase.
(c) The equivalent circuit of M1 phase. (d) The equivalent circuit of M2 phase.

 vdc ( Ra - Rb ) - (i1 - i2 )( Ra Rb )
R f 1 = i1 Rb - i2 Ra

 (1)
 R = vdc ( Ra - Rb ) - (i1 - i2 )( Ra Rb )
 f 2 i1 Ra - i2 Rb

When the DC system has GC, the capacitance value of the DC negative pole to the earth is C1,
and the capacitance value of the DC positive pole to the earth is C2. Thus, the circuits in Figure 1c,d
can be changed to those shown in Figure 2a,b. To facilitate calculation and analysis, the equivalent
Appl. Sci. 2019, 9, 2607 4 of 16

resistance of the two working modes is assumed to be what is shown in Equation (2). Figure 2a,b can
be simplified as Figure 2c,d.

 Rb R f 1 Ra R f 2
 R11 = , R12 =
 Rb + R f 1 Ra + R f 2
 (2)
 R = Ra R f 1 , R = Rb R f 2
 21 R + R 22
Rb + R f 2
 a f1

(a) (b) (c) (d)

Figure 2. Equivalent circuit of unbalanced bridge with GC. (a) The circuit of M1 phase. (b) The
circuit of M2 phase. (c) The equivalent circuit of M1 phase. (d) The equivalent circuit of M2 phase.

The following parameters are set.

 R11 R12
 X 11 = R + R , X 12 = R + R
 11 12 11 12
 (3)
X = R R
21
, X 22 = 22


21
R21 + R22 R21 + R22

According to Figure 2, the time constants of the M1 and M2 phases are defined as
(C1 + C2 ) R11 R12
τ1 = and τ 2 = (C1 + C2 ) R21 R22 . (4)
( R11 + R12 ) ( R21 + R22 )

When the two phases switch with each other, the charging process of GC belongs to the
first-order circuit full response process, and the curvilinear function Equation (5) can be obtained,
where v110, v120, v210, and v220 are the initial voltage of the full response processes of v11, v12, v21, and v22,
respectively.

 -
τ
t

v11 = vdc X 11 + (v110 - vdc X 11 )e 1

 t
 -
τ
v12 = vdc X 12 + (v120 - vdc X 12 )e 1
 t (5)
 -
τ2
 v21 = vdc X 21 + ( v210 - vdc X 21 ) e
 -
t
v22 = vdc X 22 + (v220 - vdc X 22 )e τ 2

3. New Strategy to Avoid the Impact on GC: Three-Point Climbing Algorithm

The traditional sampling method when GC exists is shown in Figure 3. In the M1 phase, v1
decreases slowly, and v2 increases slowly. Sampling continues for v1 and v2 until t11_n and t12_n,
respectively; v1 falls to stable value vdcX11, and v2 rises to stable value vdcX12. After stabilization, the
final v11 and v12 are obtained. The method switches to the M2 phase until v1 and v2 reach stable values
Appl. Sci. 2019, 9, 2607 5 of 16

vdcX21 and vdcX22, respectively, and the controller obtains the final v21 and v22. One sampling period TC
ends, and the insulation resistance values Rf2 and Rf1 are calculated by v11 = vdcX11, v12 = vdcX12, v21 =
vdcX21, and v22 = vdcX22. The resistance values of R11, R12, R21, and R22 are large, so the charging time of
the capacitor is long. Sampling and calculation should be performed after GC charging is completed;
hence, the measurement time of the traditional method is long and unable to meet the real-time
requirements of EVs. To avoid the measurement overtime caused by GC, a new insulation resistance
monitoring method, namely, three-point climbing algorithm, is proposed.

Q1

Q2
v21
vdcX21

v1
v11
vdcX11
Δt
t11_1 t11_2 t11_3 t11_4 t11_n t21_1 t21_2 t21_3 t21_4 t21_n
v12
vdcX12

v2
v22
vdcX22
Δt
t12_1 t12_2 t12_3 t12_4 t12_n t22_1 t22_2 t22_3 t22_4 t22_n
M1 M2
TC

Figure 3. Traditional sampling method.

Figure 4 shows that each phase is sampled three times, and the sampling intervals are equal.
With v11 as an example, the three sampling times are t11_1, t11_2, and t11_3; the sampling voltage values
are v11_1, v11 _2, and v11_3, respectively, and the time intervals are Δt. Similar definitions of v12, v21, and
v22 are provided to facilitate the calculation, and E1 and E2 are used to present natural exponential
function. The following equation is then obtained from (5).

 v11_1 - v11_ 2 -
Δt
v12_1 - v12_ 2 -
Δt

 = e τ1 =E1 , = e τ1 =E1
 v11_ 2 - v11_ 3 v12_ 2 - v12_ 3
 Δt Δt (6)
 v21_1 - v21_ 2 -
τ2
v22_1 - v22_ 2 -
τ2
v = e =E2 , =e =E2
 21_ 2 - v21_ 3 v22_ 2 - v22_ 3

t11_1, t12_1, t21_1, and t22_1 are set as initial times for the first-order circuit full response curve, and
the curve Equation (5) is converted into Equation (7).

v11_ 2 = vdc X 11 + (v11_1 - vdc X 11 ) E1


v12 _ 2 = vdc X 12 + (v12 _1 - vdc X 12 ) E1
 (7)
v21_ 2 = vdc X 21 + (v21_1 - vdc X 21 ) E2
v
 22_ 2 = vdc X 22 + (v22 _1 - vdc X 22 ) E2
Equation (7) is converted into Equation (8).

 v11_ 2 - v11_1 E1 v12 _ 2 - v12 _1 E1

 X 11 = , X 12 =
 vdc (1 - E 1 ) vdc (1 - E1 )
 (8)
 X = v21_ 2 - v12 _1 E2 , X = v22 _ 2 - v22 _1 E2
 21 vdc (1 - E2 )
22
vdc (1 - E2 )

Appl. Sci. 2019, 9, 2607 6 of 16

X11, X12, X21, and X22 in Equation (8) are known values calculated by sampling voltage. Equation
(9) can be obtained from Equations (2), (3), and (8), and insulation resistance values Rf1 and Rf2 can be
solved.

 Ra Rb Ra Rb
Rf 1 = R - R , Rf 2 =
Rb - Ra
 a b
X 12 - Ra X 11 - Rb
 X 12 - X 22 X 11 - X 21
 (9)
R = Ra Rb Ra Rb
,R f 2 =
 f1 Ra - Rb Rb - Ra
 X 22 - Rb X 21 - Ra
 X 12 - X 22 X 11 - X 21

Q1

Q2

v110 v21_3
v11_1 v21_2
v1 v11_2
v21_1
v11_3 v210
Δt Δt Δt Δt
t11_1 t11_2 t11_3 t21_1 t21_2 t21_3

v12_3 v220
v12_2 v22_1
v2 v22_2
v12_1
v120 v22_3

Δt Δt Δt Δt
t12_1 t12_2 t12_3 t22_1 t22_2 t22_3
M1 M2
TC

Figure 4. Proposed three-point climbing algorithm.

4. Implementation Method for Improving Accuracy

4.1. Error Analysis

In practical applications, sample resolution and voltage ripple cause sampling errors, which
affect the calculated result of insulation resistance. With the v11 of the M1 phase as an example, Δv11_1,
Δv11_2, and Δv11_3 are the sampling errors of v11_1, v11_2, and v11_3, respectively. E1R is the actual value of
E1. E1C is the measured value of E1. Considering the sampling error, the expression of E1R and E1C
according to Equation (6) is shown as

 v11_ 2 -v11_ 3
 E1R =
 v11_1 -v11_ 2
 . (10)
 E = (v11_ 2 +Δv11_ 2 ) - (v11_ 3 +Δv11_ 3 )
 1C (v +Δv ) - (v +Δv )
 11_1 11_1 11_ 2 11_ 2

ΔE1 is the error of E1 and is defined as follows:

ΔE1 = E1C − E1R . (11)

By substituting Equation (11) into Equation (10), ΔE1 can be rewritten as

(Δv11_ 2 - Δv11_ 3 ) − E1R (Δv11_1 - Δv11_ 2 )
ΔE1 = (12)
(v11_1 -v11_ 2 )+(Δv11_1 - Δv11_ 2 )
Appl. Sci. 2019, 9, 2607 7 of 16

where Δv11_1, Δv11_2, and Δv11_3 are uncontrollable components and E1R is a fixed value. (v11_1–v11_2) is
inversely proportional to ΔE1. Similarly, v12, v21, and v22 can result in the same conclusion. In the
application, the larger Δt11, Δt12, Δt21, and Δt22 are, the smaller the result error is. The larger the
difference between the Ra and Rb is, the smaller the result error is. The higher the vdc is, the smaller
the result error is.

4.2. Selection of Calculation Method

In the climbing stage, the smaller GC is, the larger the difference is in the three-point sampling
and the smaller the result error is. If GC is too small, the sampling may reach the stable stage after
climbing, the difference between the three sampled voltages will be too small, and the resulting error
will increase. The accuracy is the highest only when the three sampled voltages are in the climbing
stage of the voltage variation curve. Equation (6) shows that when the difference of the three
sampled voltages is close to zero, the calculated value of ΔE1 is nearly infinite because of signal
interference in the actual application unlike in the ideal situation. This condition seriously affects the
measurement result. A positive constant is set as Ca to determine if GC is too small by using the
following rule.

v11_ 2 -v11_ 3 < Ca


v22_ 2 -v22_ 3 < Ca
 (13)
v12_ 2 -v12_ 3 > −Ca
v -v
 21_ 2 21_ 3 > −Ca
If the three sampled voltages satisfy Equation (13), then GC is too small to be non-negligible,
and the traditional sampling method is adopted. If the three sampled voltages do not satisfy
Equation (13), then GC is non-negligible, and the proposed three-point climbing algorithm is
adopted.

4.3. Filter of E1 and E2

Δt is an invariant constant; τ1 and τ2 vary with GC, so E1 and E2 are also variations. Counter
k is increased once every measurement period TC, as shown in Figure 5. To make kth measurements
E1 and E2 close to the actual value, the average value of M1 and M2 phases are taken according to
Equation (6). E1(k) and E2(k) can be rewritten as

 v11_1 (k ) - v11_ 2 (k ) v12_1 (k ) - v12_ 2 (k )

 E1 (k ) = ( + )/2
 v11_ 2 (k ) - v11_ 3 (k ) v12_ 2 (k ) - v12_ 3 (k )
 . (14)
 E (k ) = ( v21_1 (k ) - v21_ 2 (k ) + v22_1 (k ) - v22_ 2 (k ) ) / 2
 2 v21_ 2 (k ) - v21_ 3 (k ) v22_ 2 (k ) - v22_ 3 (k )


1
vdc
2

Figure 5. Waveform of the bridge voltage.

Appl. Sci. 2019, 9, 2607 8 of 16

The estimated value of the kth E1 and E2 is set as Eˆ1 (k ) and Eˆ 2 (k ) , respectively, which can be
obtained by the following first-order filter, where A is a filter coefficient that satisfies 0 < A < 1.

 Eˆ1 (k ) = AEˆ1 (k − 1) + (1 − A) E1 (k )
 (15)
 Eˆ 2 (k ) = AEˆ 2 (k − 1) + (1 − A) E2 (k )

4.4. Correction of Sampled Value

The actual sampled voltage value shows a certain deviation from the expected value. The red
sampled point in Figure 6 shows that the exponential function curve cannot be formed. The result
calculated by the sampled value must be a large error. Therefore, to obtain satisfactory results, the
sampled values must be corrected with a loop iterative correction method. vˆ11_1 (i) , vˆ11_ 2 (i) , vˆ11_ 3 (i) ,
vˆ12 _1 (i) , vˆ12 _ 2 (i) , vˆ12 _ 3 (i) , vˆ21_1 (i) , vˆ21_ 2 (i) , vˆ21_ 3 (i) , vˆ22 _1 (i) , vˆ22 _ 2 (i) , and v22 _ 3 (i) are set as the ith
correction voltage values. After the 12 voltages of v11, v12, v21, and v22 are sampled completely, the
counter is set as i = 0. The 12 voltage values are substituted into the following equation, and the
sampled value is used as the initial correction value.

vˆ11_1 (0) = v11_1 vˆ12 _1 (0) = v12_1 vˆ21_1 (0) = v21_1 vˆ22 _1 (0) = v22_1
   
vˆ11_ 2 (0) = v11_ 2 , vˆ12 _ 2 (0) = v12_ 2 vˆ21_ 2 (0) = v21_ 2 , vˆ22 _ 2 (0) = v22_ 2 (16)
ˆ   
v11_ 3 (0) = v11_ 3 vˆ12 _ 3 (0) = v12_ 3 vˆ21_ 3 (0) = v21_ 3 vˆ22 _ 3 (0) = v22_ 3

v11_1
v̂11_1
v̂11_ 2
v11_ 3
v11_ 2
v̂11_ 3

Figure 6. Actual sampling point.

The three sampled voltages of each group cannot form an exponential curve of Eˆ1 (k ) due to
the measurement error and ripple. To form the desired exponential curve, each voltage value can be
estimated by the two other voltage values. v11_1 (i) , v11_ 2 (i) , v11_ 3 (i) , v12 _1 (i) , v12_ 2 (i) , v12_3 (i) , v21_1 (i) ,
v21_ 2 (i) , v21_ 3 (i) , v22_1 (i) , v22 _ 2 (i) , and v22 _ 3 (i) are set as the estimated voltage values. The estimated
method can be derived from the following equation according to Equation (6).

v11_1 (i ) = [vˆ11_ 2 (i ) − vˆ11_ 3 (i )]Eˆ1 (k )−1 + vˆ11_ 2 (i ) v12 _1 (i ) = [vˆ12 _ 2 (i ) − vˆ12 _ 3 (i )]Eˆ1 (k ) −1 + vˆ12 _
 
v11_ 2 (i ) = [vˆ11_1 (i ) Eˆ1 (k ) + vˆ11_ 3 (i )] / (1 + Eˆ1 (k )), v12 _ 2 (i ) = [vˆ12 _1 (i ) Eˆ1 (k ) + vˆ12 _ 3 (i )] / (1 + Eˆ
 
v11_ 3 (i ) = vˆ11_ 2 (i ) − [vˆ11_1 (i ) − vˆ11_ 2 (i )]Eˆ1 (k ) v12 _ 3 (i ) = vˆ12 _ 2 (i ) − [vˆ12 _1 (i ) − vˆ12 _ 2 (i )]Eˆ1 (k
v21_1 (i ) = [vˆ21_ 2 (i ) − vˆ21_ 3 (i )]Eˆ 2 (k ) −1 + vˆ21_ 2 (i ) v22 _1 (i ) = [vˆ22 _ 2 (i ) − vˆ22 _ 3 (i )]Eˆ 2 (k )−1 + vˆ2 (17)
 
v21_ 2 (i ) = [vˆ21_1 (i ) Eˆ 2 (k ) + vˆ21_ 3 (i )] / (1 + Eˆ 2 (k )) , v22 _ 2 (i ) = [vˆ22 _1 (i ) Eˆ 2 (k ) + vˆ22 _ 3 (i )] / (1 +
 ˆ  ˆ
v21_ 3 (i ) = v21_ 2 (i ) − [v21_1 (i ) − v21_ 2 (i )]E2 (k ) v22 _ 3 (i ) = v22 _ 2 (i ) − [v22 _1 (i) − v22 _ 2 (i )]E2
 ˆ ˆ ˆ  ˆ ˆ ˆ

The estimated value comparison rule is shown as

Appl. Sci. 2019, 9, 2607 9 of 16

 vˆ11_1 (i ) − v11_1 (i ) < D, vˆ11_ 2 (i) − v11_ 2 (i ) < D, vˆ11_ 3 (i ) − v11_ 3 (i) < D


 vˆ12 _1 (i) − v12 _1 (i) < D, vˆ12 _ 2 (i) − v12 _ 2 (i) < D, vˆ12 _ 3 (i) − v12 _ 3 (i) < D
. (18)
 vˆ21_1 (i ) − v21_1 (i) < D, vˆ21_ 2 (i ) − v21_ 2 (i ) < D, vˆ21_ 3 (i) − v21_ 3 (i) < D


 vˆ22 _1 (i) − v22 _1 (i ) < D, vˆ22 _ 2 (i) − v22 _ 2 (i) < D, vˆ22 _ 3 (i ) − v22 _ 3 (i ) < D

When Equation (18) is satisfied, the difference between the estimated value and the correction
value is small, and the ith correction value is applied as the final correction value. Otherwise, the
counter i is increased by 1, and further correction is be conducted as follows:

vˆ11_1 (i + 1) = v11_1 (i ) + B[vˆ11_1 (i ) − v11_1 (i )] vˆ12 _1 (i + 1) = v12 _1 (i ) + B[vˆ12 _1 (i ) − v12 _1 (i)]

 
vˆ11_ 2 (i + 1) = v11_ 2 (i ) + B[vˆ11_ 2 (i ) − v11_ 2 (i )], vˆ12 _ 2 (i + 1) = v12 _ 2 (i ) + B[vˆ12 _ 2 (i ) − v12 _ 2 (i )]
ˆ 
v11_ 3 (i + 1) = v11_ 3 (i ) + B[vˆ11_ 3 (i ) − v11_ 3 (i )] vˆ12 _ 3 (i + 1) = v12 _ 3 (i ) + B[vˆ12 _ 3 (i ) − v12 _ 3 (i )]
vˆ21_1 (i + 1) = v21_1 (i ) + B[vˆ21_1 (i ) − v21_1 (i )] vˆ22 _1 (i + 1) = v22 _1 (i ) + B[vˆ22 _1 (i ) − v22 _1 (i )] (19)
 
vˆ21_ 2 (i + 1) = v21_ 2 (i ) + B[vˆ21_ 2 (i ) − v21_ 2 (i )] , vˆ22 _ 2 (i + 1) = v22 _ 2 (i ) + B[vˆ22 _ 2 (i ) − v22 _ 2 (i )]
ˆ 
v21_ 3 (i + 1) = v21_ 3 (i ) + B[vˆ21_ 3 (i ) − v21_ 3 (i )] vˆ22 _ 3 (i + 1) = v22 _ 3 (i ) + B[vˆ22 _ 3 (i ) − v22 _ 3 (i )]

where B is the correction factor that satisfies 0 < B < 1. Then, the results of Equation (19) are
substituted into Equation (17). This method cycles back and forth until the difference between the
estimated and correction values satisfies Equation (18). The cycle is then stopped, and the final
correction value is outputted.
The overall software flow chart of the method is shown in Figure 7.
Appl. Sci. 2019, 9, 2607 10 of 16

Start

Open M1 phase Open M2 phase

Delay Delay

Sample v11，v12 Sample v21，v22

Delay Δt Delay Δt

Three sets of data Three sets of data

N were sampled? N were sampled?

Y Y

Eqn.(14) N Satisfy the

Eqn.(15) Eqn.(13)?
Eqn.(16)
Y

k++ i++ Get Rf1、Rf2 according

Eqn.(17) to Eqn.(1)

Eqn.(19)

Satisfy the
N Eqn.(18)?

Eqn.(6)
Eqn.(8)

Get Rf1、Rf2 according

to Eqn.(9)

Figure 7. Software flow chart of the proposed method.

5. Comparative Study of Experimental Data

The DC-IM device with the unbalanced electric bridge method is created, and the schematic
overview of application is shown in Figure 8. The MCU performs the switch Q1 and Q2, sample the
voltage values of v1 and v2 stored in memory, then calculate the Rf1 and Rf2, and output the result to
the computer. The experiment table is shown in Figure 9. It includes the display interface, DC-IM
device, voltage regulating device, insulation resistance selection switch, and GC selection switch.
The DC-IM controller uses a PIC18F4580 single-chip microcomputer. The monitoring period is 0.2 s;
that is, the switch action occurs every 0.1 s, so the sampling time should satisfy (t1 + 2Δt) < 0.1 s. The
positive grounding resistance is Rf1 = 1000 kΩ, the negative grounding resistance is Rf2 = 300 kΩ, the
first sampling time t1 = 0.01 s, the bridge resistance is Ra = 1000 kΩ, and Rb = 200 kΩ. The GC value is
CY = C1 = C2. The grounding current waveform corresponding to different GCs is shown in Figure
10. The waveform of CY = 0.1 μF can be stabilized in a half cycle. The larger the value of CY is, the
closer the waveform is to the triangle wave. Therefore, the traditional unbalanced bridge sampling
method is used when CY < 0.1 μF, and the three-point climbing algorithm is used when CY > 0.1 μF.
The calculation results are compared by changing the different parameters, and relative error
(RE%) is determined as
Appl. Sci. 2019, 9, 2607 11 of 16

RE% = | measured value − actual value |/measured value. (20)

Different parameters are applied in the proposed method. (1) DC voltage vdc = 800 V, GC value
CY = 0.1 μF, and the sampling time interval Δt is changed; the results are shown in Figure 11. The
larger the sampling time interval is, the higher accuracy is. (2) DC voltage vdc = 800 V, sampling
time interval Δt = 0.04 s, and the value of CY is changed; the results are shown in Figure 12. The
smaller CY is, the higher accuracy is. (3) Sampling interval Δt = 0.04 s, CY = 0.1 μF, and DC voltage
vdc is changed; the results are shown in Figure 13. The larger the DC voltage is, the higher accuracy
is. When the DC voltage drops to below 200 V, the measurement accuracy is greatly reduced. (4)
Under the premise that the parallel value of bridge resistance Ra || Rb is constant and the difference
between Ra and Rb is changed; the results are shown in Table 2. The larger the difference between Ra
and Rb is, the higher accuracy is. Time interval Δt increases, CY decreases, DC voltage vdc increases,
and the difference between Ra and Rb increases. These factors make the sampled voltage difference
larger, which will reduce the error of the final results.
The proposed method is compared with the traditional method to verify the availability and
superiority of the former. The monitoring time and relative error of the two methods are shown in
Tables 3–5. The relative error is the larger one between Rf1 and Rf2. Table 3 is the data at vdc = 800 V,
Rf1 = 1000 kΩ, Rf2 = 300 kΩ; Table 4 is the data at vdc = 400 V, Rf1 = 1000 kΩ, Rf2 = 300 kΩ; Table 5 is the
data at vdc = 800 V, Rf1 = 100 kΩ, Rf2 = 100 kΩ. The traditional method needs to increase the
monitoring time with a large value of GC because it should have a stable sample value after charging
the GC. The proposed method has a fixed monitoring time due to the fixed three-point sampling.
When GC is large and the proposed method is applied, the error is similar to using the traditional
method. When GC is small, such as 10 nF in the table, and the traditional unbalanced bridge
calculation method is applied automatically, the calculation results of two methods are almost the
same. Overall, the experimental results are consistent with the theoretical conclusion.

DC+ DC-

R1 R1
Q2 Q1

R2 R2
v2 v1
A A
D D
perform

upper computer v12, v22 v11, v21

memory
MCU
USB calculation

output Rf1, Rf2

RS485
converter

Figure 8. The schematic overview of application.

Appl. Sci. 2019, 9, 2607 12 of 16

display interface
DC voltage capacitance selection

insulation resistance
selection

insulation monitoring

Figure 9. Display of the experiment table.

(a) CY = 0.1 μF

(b) CY = 0.2 μF
Appl. Sci. 2019, 9, 2607 13 of 16

(c) CY = 0.4 μF
Figure 10. Waveform of grounding current with different values of CY. (a) CY = 0.1 μF. (b) CY = 0.2 μ.
(c) CY = 0.4 μF.

5.00% Rf 1
4.00% Rf 2
3.00%

2.00%

1.00%

0.00%
0.01s 0.02s 0.03s 0.04s

Figure 11. RE% with different sampling intervals.

8.00% Rf 1
7.00%
6.00% Rf 2
5.00%
4.00%
3.00%
2.00%
1.00%
0.00%
10nF 0.1μF 1μF 2μF 3μF 4μF

Figure 12. RE% with different values of CY.

20.00% Rf 1
15.00% Rf 2

10.00%

5.00%

0.00%
1000V 900V 800V 700V 600V 500V 400V 300V 200V 100V

Figure 13. RE% with different DC voltages.

Appl. Sci. 2019, 9, 2607 14 of 16

Table 2. Data results with different bridge resistors.

Ra (kΩ) Rb (kΩ) Rf1 (kΩ) RE% of Rf1 Rf2 (kΩ) RE% of Rf2
1000 200 1010 1.0% 302 0.6%
800 250 1014 1.4% 305 1.6%
600 300 1020 2.0% 309 3.0%
500 350 1045 4.5% 316 5.3%

Table 3. Comparison of monitoring data in vdc = 800 V, Rf1 = 1000 kΩ, Rf2 = 300 kΩ.

Traditional Method Proposed Method

CY Monitoring Time RE% Monitoring Time RE%
10 nF 0.2s 0.8% 0.2s 0.8%
0.1 μF 0.3s 1.0% 0.2s 1.2%
1 μF 1.35s 2.4% 0.2s 1.8%
2 μF 2.51s 3.5% 0.2s 3.7%
3 μF 3.67s 5% 0.2s 5.3%
4 μF 4.83s 6.6% 0.2s 7.3%

Table 4. Comparison of monitoring data in vdc = 400 V, Rf1 = 1000 kΩ, Rf2 = 300 kΩ.

Traditional Method Proposed Method

CY Monitoring Time RE% Monitoring Time RE%
10 nF 0.2s 1.3% 0.2s 1.3%
0.1 μF 0.29s 1.7% 0.2s 2.8%
1 μF 0.34s 3.2% 0.2s 4.1%
2 μF 2.49s 5.8% 0.2s 6.2%
3 μF 3.64s 8.5% 0.2s 9%
4 μF 4.79s 11% 0.2s 12%

Table 5. Comparison of monitoring data in vdc = 800 V, Rf1 = 100 kΩ, Rf2 = 100 kΩ.

Traditional Method Proposed Method

CY Monitoring Time RE% Monitoring Time RE%
10 nF 0.2s 0.7% 0.2s 0.7%
0.1 μF 0.23s 1.0% 0.2s 0.9%
1 μF 0.65s 2.2% 0.2s 1.6%
2 μF 1.11s 3.4% 0.2s 2.4%
3 μF 1.57s 4.7% 0.2s 4.1%
4 μF 2s 6.3% 0.2s 5.2%

6. Conclusion
For the DC-IM circuit of EVs, the traditional unbalanced electric bridge method switches the
positive and negative bridge resistances and calculates the insulation resistance value by sampling
the positive and negative bridge voltages. However, when the DC positive and negative poles have
GC, the bridge voltages must be sampled after the capacitor is charged completely; thus, the
measurement time is very long. This study proposes a novel method of DC-IM using a three-point
climbing algorithm. The insulation resistance can be calculated by sampling the voltage of the
positive and negative bridges three times and keeping the sampling interval equal. Moreover, the
method filters and automatically corrects the three sampling voltages, which can improve the
accuracy of the calculation results. Combined with experimental data, the following conclusions can
be drawn: (1) The advantage of proposed method can perform a faster time and maintain a constant
monitoring period compared with the traditional method. (2) The restriction of proposed method
Appl. Sci. 2019, 9, 2607 15 of 16

only apply in larger GC situation. If GC is small, the traditional method could be used. (3) The
characteristics of proposed method: Increasing the sampling interval, increasing the difference
between Ra and Rb, increasing DC voltage vdc, all make the results more accurate. Overall, the
proposed method can be applied to some practical industrial applications. The future work is to
study how to set the constant Ca to determine which method to use or find a different rule to
generally judge the value of GC, and study a method to reduce error when vdc is constant changing.
Author Contributions: J.D. and T.Q.Z. provided the method and solution; Y.Y. and H.Z. performed the
experiments; J.D. and Y.Y. analyzed the data; J.D. and Y.Z. wrote the paper; T.Q.Z. and H.L. checked paper.

Funding: This research received no external funding.

Conflicts of Interest: The authors declare no conflict of interest.

References
1. Hou, Y.P.; Zhou, W.; Shen, C.Y. Experimental investigation of gas-tightness and electrical insulation of
fuel cell stack under strengthened road vibrating conditions. Int. J. Hydrogen Energy 2011, 36, 13763–13768.
2. Gyftakis, K.N.; Sumislawska, M.; Kavanagh, D.F.; Howey, D.A.; McCulloch, M.D. Dielectric
Characteristics of Electric Vehicle Traction Motor Winding Insulation Under Thermal Aging. IEEE Trans.
Ind. Appl. 2016, 52, 1398–1404.
3. Dannier, A. Overview of Main Electric Subsystems of Zero-Emission Vehicles. Propuls. Syst. 2019, 1–31,
doi:10.5772/intechopen.82876.
4. Vahedipour-Dahraie, M.; Rashidizaheh-Kermani, H.; Najafi, H.R.; Anvari-Moghaddam, A.; Guerrero, J.M.
Coordination of EVs Participation for Load Frequency Control in Isolated Microgrids. Appl. Sci. 2017, 7,
539.
5. Rafique, M.K.; Khan, S.U.; Zaman, M.S.U.; Mehmood, K.K.; Haider, Z.M.; Ali Bukhar, S.B.; Kim, C. An
Intelligent Hybrid Energy Management System for a Smart House Considering Bidirectional Power Flow
and Various EV Charging Techniques. Appl. Sci. 2019, 9, 1658.
6. Kim, J.; Kim, H.; Cho, Y.; Kim, H.; Cho, J. Application of a DC Distribution System in Korea: A Case
Study of the LVDC Project. Appl. Sci. 2019, 9, 1074.
7. Holger Potdevin, D.-I. Insulation Monitoring in High Voltage Systems for Hybrid and Electric Vehicles.
High Voltage Safety. ATZ Elektron. 2009, 4, 28–31.
8. Pan, X.F.; Rinkleff, T.; Willmann, B.; Vick, R. PSpice simulation of surge testing for electrical vehicles. In
Proceedings of the 2012 International Symposium on Electromagnetic Compatibility (EMC EUROPE),
Rome, Italy, 17–21 September 2012.
9. Liu, Y.-C.; Lin, C.-Y. Insulation fault detection circuit for ungrounded DC power supply systems. In
Proceedings of the 2012 IEEE Sensors, Taipei, Taiwan, 28–31 October 2012.
10. Zhao, C.Y.; Jia, X.F.; Hao, Z.F. The new method of monitoring DC system insulation on-line. In
Proceedings of the 27th Annual Conference of the IEEE Industrial Electronics Society, IECON’01, Denver,
CO, USA, 29 November–2 December 2001.
11. Jiang, J.S.; Ji, H. Study of Insulation Monitoring Device for DC System Based on Multi-switch
Combination. In Proceedings of the Second International Symposium on Computational Intelligence and
Design, ISCID’09, Changsha, China, 12–14 December 2009.
12. Li, J.X.; Wu, Z.Q.; Fan, Y.Q.; Wang, Y.Y.; Jiang, J.C. Research on Insulation Resistance On-Line Monitoring
for Electric Vehicle. In Proceedings of the 2005 International Conference on Electrical Machines and
Systems, Nanjing, China, 27–29 September 2005.
13. Wei, X.Y.; Bi, L.; Sun, Z.C. A method of insulation failure detection on electric vehicle based on FPGA. In
Proceedings of the 2008 IEEE Vehicle Power and Propulsion Conference, VPPC’08, Harbin, China, 3–5
September 2008.
14. Zheng, M.X. The Design of Insulation Monitoring Device for Electric Vehicle’s Battery Pack. Adv. Mater.
Res. 2013, 718–720, 1422–1428.
15. Bao, Y.; Jiang, J.C.; Zhang, W.G.; Wang, J.Y.; Wen, J.P. Novel On-line Insulation Supervising Method for
DC System. High Volt. Eng. 2011, 37, 333–337.
16. Wu, Z.J.; Wang, L.F. A novel insulation resistance monitoring device for Hybrid Electric Vehicle. In
Proceedings of the 2008 IEEE Vehicle Power and Propulsion Conference, VPPC’08, Harbin, China, 3–5
September 2008.
Appl. Sci. 2019, 9, 2607 16 of 16

17. Neti, P.; Grubic, S. Online broadband insulation spectroscopy of induction machines using signal
injection. IEEE Trans. Ind. Appl. 2017, 53, 1054–1062.
18. Chiang, Y.-H.; Sean, W.-Y.; Huang, C.-Y.; Chiang Hsieh, L.-H. Adaptive control for estimating insulation
resistance of high-voltage battery system in electric vehicles. Environ. Prog. Sustain. Energy 2017, 36, 1882–
1887.
19. Li, D.H.; Jia, W. Application of Chaos Theory in Grounaing Fault Detection of DC Power Supply System.
High Volt. Eng. 2005, 31, 79–82.
20. Li, D.H.; Shi, L.T. An Overall Scheme for Ground Fault Detection in DC Systems of Power Plants and
Substations. Power Syst. Technol. 2005, 29, 56–59.
21. Song, C.X.; Shao, Y.L.; Song, S.X.; Peng, S.L.; Zhou, F.; Chang, C.; Wang, D. Insulation Resistance
Monitoring Algorithm for Battery Pack in Electric Vehicle Based on Extended Kalman Filtering. Energies
2017, 10, 714.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).