GROUND FAULT DETECTION FOR FLEXIBLE
HIGH VOLTAGE POWER SYSTEMS
By
Aravind Mathsyaraja

A THESIS
Submitted to
Michigan State University
In partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE
Electrical Engineering
2010

ABSTRACT
GROUND FAULT DETECTION FOR FLEXIBLE HIGH VOLTAGE
POWER SYSTEMS
By
Aravind Mathsyaraja
Hybrid Electric Vehicles (HEV) in the consumer and commercial vehicle sectors
have seen tremendous technological advancement in the last decade. The commercial
vehicle industry in particular has benefited significantly from the hybridization of the
propulsion system, and unlike the automotive segment, serves a wide variety of
functions. This diversity in vehicle applications drives the necessity for high voltage
power systems to be flexible in nature, allowing them to adapt to different vehicle
architectures while performing the intended function. As a result, diagnostic modules
within the high voltage power system, such as ground fault detection circuits, are being
required to operate robustly in a high voltage power system that is exposed to electrical
noise and significant variation in common mode impedance characteristics.
This paper explores four different ground fault detection methodologies that exist
today in the industry, and evaluates their performance in a flexible high voltage power
system. A thorough comparison of these technologies is performed based on results from
seven distinct test cases, followed by a recommendation for the ideal ground fault
detection system.

To my parents, Haragopal and Vijaya Lakshmi Mathsyaraja
Whom I love and respect unconditionally

iii

ACKNOWLEDGMENTS

Writing this thesis has been a tremendous learning experience for me and has
offered me a sense of true accomplishment. It would not, however, have been possible
without the support, encouragement, and guidance provided by many at the university,
work, and home.
I would like to first and foremost offer my sincere gratitude to my manager, Ben
Saltsman, for (perhaps unknowingly) helping me make a firm and final decision to
complete my thesis by December 2010. Reminding me to be accountable to my original
plan, he allowed me to act upon the decision and see the thesis through to completion.
I am grateful to Dr. Xiaosong Kang for his valuable suggestions and guidance in
selecting a topic for research, and to my graduate advisor Dr. Fang Zeng Peng for his
approval of my work, understanding, and consistent encouragement in spite of the delays
caused by my relocation in 2010.
I am blessed to be surrounded by supportive colleagues who were always open to
discussions regarding my research. One person in particular who has guided me in too
many ways to be able to mention in one page is Thomas Stoltz, for his thought provoking
insights and feedback form the basis of my approach to writing the thesis.
Finally, the journey towards completion would not have been nearly as enjoyable
or fulfilling if not for the support and encouragement of my dear friends Sumeet & Ketki
Oswal, and the love of my beautiful wife Shweta.

iv

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... vii
LIST OF FIGURES ....................................................................................................... viii
1 Introduction ...................................................................................................................... 1
2 Failure Modes & Root Causes ......................................................................................... 4
2.1 Resistive Leak Fault on HVDC Bus ......................................................................... 4
2.1.1 Wire Chafing ...................................................................................................... 4
2.1.2 Aging of Components......................................................................................... 6
2.1.3 Contaminant Intrusion ........................................................................................ 6
2.1.4 Manufacturing Defect ......................................................................................... 7
2.1.5 Maintenance worker incident ............................................................................. 7
2.2 Alien Component Introduced to System ................................................................... 8
2.2.1 Excessive filtering to ground .............................................................................. 8
2.2.2 Auxiliary Loads with Pre-existing Faults ......................................................... 10
2.3 Leak Fault on the HVAC Bus ................................................................................. 10
2.3.1 Motor winding insulation fault ......................................................................... 11
3 Effects of Ground Faults ................................................................................................ 13
3.1 Interference with other Circuits............................................................................... 13
3.2 Potential Shock Hazards (AC and DC current) ....................................................... 13
3.3 Damage of Auxiliary Filter Circuits........................................................................ 17
4 Survey and Analysis of Existing Technologies ............................................................. 19
4.1 Definition of Flexible HEV Architecture ................................................................ 19
4.2 Nissan Motor – Ground Fault Detector for Vehicle ................................................ 21
4.2.1 Technology Overview ...................................................................................... 21
4.2.2 Simulation and Analysis ................................................................................... 27
4.3 Maxim Integrated Products – Fault Detecting Method for Detecting Leak Paths
between Power Sources and Chassis ............................................................................. 46
4.3.1 Technology Overview ...................................................................................... 46
4.3.2 Simulation and Analysis ................................................................................... 52
4.4 Lear Corporation – Ground Fault Detection System for Vehicles with High Voltage
Power Net ...................................................................................................................... 61
4.4.1 Technology Overview ...................................................................................... 61
4.4.2 Simulation & Analysis ..................................................................................... 65

v

4.5 Caterpillar Inc – Systems and Methods for Electrical Leakage Detection.............. 76
4.5.1 Technology Overview ...................................................................................... 76
4.5.2 Simulation and Analysis ................................................................................... 83
5 Comparison of Technologies ......................................................................................... 97
5.1 Criteria for Analysis and Weighting Factor ............................................................ 97
5.1.1 Detection time .................................................................................................. 98
5.1.2 Type of Fault .................................................................................................... 99
5.1.3 Mode of Detection .......................................................................................... 100
5.1.4 Flexibility of Design ....................................................................................... 101
5.1.5 Susceptibility to common noise sources......................................................... 102
5.1.6 Cost & Complexity ......................................................................................... 103
5.2 System Benchmarking........................................................................................... 104
5.3 Pugh Analysis ........................................................................................................ 111
6 Conclusion ................................................................................................................... 112
BIBLIOGRAPHY ......................................................................................................... 116

vi

LIST OF TABLES

Table 1: Required trip threshold for charger circuit interrupting devices [10] ................. 17
Table 2: Definition of ratings for criteria 1....................................................................... 98
Table 3: Definition of ratings for criteria 2....................................................................... 99
Table 4: Definition of ratings for criteria 3..................................................................... 100
Table 5: Definition of ratings for criteria 4..................................................................... 101
Table 6: Definition of ratings for criteria 5..................................................................... 103
Table 7: Definition of ratings for criteria 6..................................................................... 104
Table 8: Pugh Matrix ...................................................................................................... 111

vii

LIST OF FIGURES

Figure 1: Typical HV Power System on a Hybrid Electric Vehicle ................................... 5
Figure 2: Differential Mode and Common Mode filters [5] ............................................... 9
Figure 3: Equivalent circuit of traction inverter and induction motor in HEV [8] ........... 12
Figure 4: Speed and Torque during phase to ground fault [7] .......................................... 12
Figure 5: Physiological effects of electricity [9]............................................................... 15
Figure 6: Equivalent Human Body Circuit Model [10] .................................................... 16
Figure 7: Effect of back EMF on EMI filter ..................................................................... 18
Figure 8: Flexible Hybrid Electric Vehicle architecture [12] ........................................... 20
Figure 9: Schematic overview of method 1 [13] .............................................................. 22
Figure 10: Change in impedance for capacitive vs. resistive ground faults [13] .............. 25
Figure 11: Response of circuit after decrease in frequency [13] ...................................... 26
Figure 12: Response of circuit after increase in frequency [13] ....................................... 27
Figure 13: High voltage power system circuit model ....................................................... 31
Figure 14: Simulation results for test case 1 of method 1................................................. 32
Figure 15: Simulation results for test case 2 of method 1................................................. 33
Figure 16: DC ripple current in a non-insulated single phase inverter system ................. 34
Figure 17: Equivalent circuit of AC and DC common mode bus ..................................... 34
Figure 18: Simulation results for test case 3 of method 1................................................. 36
Figure 19: Simulation results for test case 4 of method 1................................................. 38
Figure 20: Simulation results for test case 5 of method 1................................................. 40
Figure 21: Simulation results for test case 6(a) of method 1 ............................................ 42

viii

Figure 22: Simulation results for test case 6(b) of method 1 ............................................ 43
Figure 23: Variation in the HV DC bus voltage during discharge cycle .......................... 44
Figure 24: Simulation results for test case 7 of method 1................................................. 45
Figure 25: Schematic overview of method 2 [18] ............................................................ 46
Figure 26: Response of I1 and I2 in a no fault condition [18] ........................................... 49
Figure 27: Response of I1 and I2 in a true fault condition [18] ......................................... 49
Figure 28: Response of I1 and I2 in a low impedance condition [18] ............................... 50
Figure 29: Method 2 leak detection circuit model ............................................................ 52
Figure 30: Simulation results for test case 1 of method 2................................................. 54
Figure 31: Simulation results for test case 2 of method 2................................................. 56
Figure 32: Zoomed simulation results for test case 2 of method 2 ................................... 56
Figure 33: Simulation results for test case 3 of method 2................................................. 57
Figure 34: Simulation results for test case 4 of method 2................................................. 58
Figure 35: Simulation results for test case 5 of method 2................................................. 59
Figure 36: Simulation results for test case 6 of method 2................................................. 60
Figure 37: Schematic overview of method 3 [19] ............................................................ 62
Figure 38: Equivalent circuit of common mode bus [19] ................................................. 63
Figure 39: Sequence of switch operation for method 3 [19] ............................................ 65
Figure 40: Leak detection circuit model for method 3 ..................................................... 66
Figure 41: Sequencing of switches for simulation............................................................ 68
Figure 42: Simulation results for test case 1 of method 3................................................. 69
Figure 43: Simulation results for test case 2 of method 3................................................. 71
Figure 44: Simulation results for test case 3 of method 3................................................. 72

ix

Figure 45: Simulation results for test case 4 of method 3................................................. 73
Figure 46: Simulation results for test case 5 of method 3................................................. 74
Figure 47: Common mode bus voltage as a result of method 3........................................ 75
Figure 48: schematic overview of typical Wheatstone bridge method [21] ..................... 76
Figure 49: Partial equivalent circuit of Wheatstone bridge [21]....................................... 78
Figure 50: Schematic overview of method 4 [20] ............................................................ 79
Figure 51: Leak detection circuit model for method 4 ..................................................... 83
Figure 52: Simulation results for test case 1 of method 4................................................. 84
Figure 53: Simulation results for test case 2 of method 4................................................. 86
Figure 54: Simulation results for test case 3 of method 4................................................. 87
Figure 55: Simulation results for test case 4 of method 4................................................. 88
Figure 56: Simulation results for test case 5 of method 4................................................. 89
Figure 57: Simulation results for test case 6 of method 4................................................. 90
Figure 58: Simulation results for test case 7 of method 4................................................. 93
Figure 59: Circuit improvement for method 4 .................................................................. 94
Figure 60: High frequency Common mode noise in PWM inverter (100V/div) [23] .... 102

x

CHAPTER 1
1 Introduction
Hybrid Electric Vehicles (HEV) in the consumer and commercial vehicle sectors
have seen tremendous technological advancement in the last decade. In the wake of
depleting fuel reserves, increasing fuel costs, and stringent emissions restrictions by
governments worldwide, many vehicle manufacturers and OEM suppliers are working to
reduce cost and increase reliability of all the components that form a hybrid vehicle,
which uses two or more energy sources for propulsion (typically gasoline/ diesel, and
battery supplied electrical energy).
Depending on the manufacturer and the application type, a HEV can be
constructed and configured in a variety of ways. In a typical vehicle, a High Voltage
battery pack usually in the range of a 100 to 1000 Volts DC, is used to power an inverter
that converts DC power to three phase AC power to drive a traction motor. The High
Voltage battery pack is also commonly used to power other electrical auxiliary devices,
such as a DC-DC converter for stepping down voltage to the vehicle’s low voltage (1214V) power net, or an auxiliary inverter unit for converting High Voltage DC to
120VAC. The power is often transmitted over High Voltage lines that could be accessible
to the user, and are generally shielded to shunt electrical noise to the chassis of the
vehicle.
The integrity of this high voltage system is critical to the reliability and safety of
the vehicle. Typically, the vehicle chassis will be electrically isolated from the terminals

1

or conductors of the high voltage system. Under normal conditions, leakage currents in
the order of micro amps exist between the conductors of the high voltage bus and the
frame of the vehicle. However, as these complex systems are housed in locations that
may observe exposure to harsh environments with rapidly changing temperature
conditions, as well as severe vibration, there exists a potential for gradual or relatively
instantaneous isolation breakdown; depending on the failure mode. This condition is
hazardous to users and personnel of the vehicle, and sometimes the vehicle itself.
This phenomenon of isolation breakdown is intrinsic to the nature of machines
and electronics, and has been acknowledged by the industry for many decades. Several
methodologies have been invented to identify ground faults within power systems and
react to prevent potential failures or hazards, such as Ground Fault Circuit Interrupters
(GFCIs). Especially in the last 5 years, many ground fault detection systems have been
created specifically for hybrid electric vehicle power systems, as this technology
continues to advance and penetrate the market. Yet, in spite of all the development that
has occurred in this field, few systems exist that are generic in nature but sufficiently
robust to be able to accurately identify low impedance to ground conditions on flexible
power systems, where the components that make up the system are interchangeable and
hence may vary significantly in design. This is particularly true for the commercial
hybrid vehicle market, which has many differences in comparison to the passenger
vehicle market.
Whereas passenger hybrid vehicles are designed for the single purpose of
transportation of people, with mostly similar duty cycles and environments, commercial

2

hybrid vehicles vary tremendously in the utility of the vehicle, and in the environments
within which they operate. In addition, mass production quantities of passenger hybrid
vehicles are far higher in comparison to those of commercial hybrid vehicles. With
applications ranging from city delivery, utility, telecommunications, and mining, to
refuse, goods transportation, mass public transit systems, and many more – the hybrid
power systems that have to be integrated into these vehicles will significantly differ in
topology and design. It has therefore become critical to design hybrid power systems that
maintain a standard framework but at the same time offer the ability to be modified as per
the requirements of the application; thence the introduction of flexible hybrid power
systems, and the need for a ground fault detection system that will function effectively
irrespective of the devices that are connected to the High Voltage Bus.
The first part of this thesis will focus on the details of hybrid vehicular ground
faults, discussing their root causes and effects. The second part, which contributes to the
primary intent of this thesis, will dive into an analysis of the various technologies that
exist today for achieving ground fault detection, and demonstrate that a ground fault
detection method for flexible hybrid power systems can be developed based on these
existing technologies. Close attention will be paid to the various attributes and
characteristics of these concepts, and eventually their strengths and weaknesses. As a
result of this analysis, a few recommendations for future work will be presented that will
lay the groundwork for the development of a potentially powerful and versatile diagnostic
system.

3

CHAPTER 2
2 Failure Modes & Root Causes
Isolation breakdown between High Voltage conductors and the chassis of the
vehicle can occur in several locations within the system, and may have been caused by
one or more of a wide range of fault conditions. In a Hybrid Electric Vehicle, a ground
fault can occur primarily between the High Voltage DC (HVDC) bus and chassis, or the
High Voltage AC (HVAC) bus and chassis. The following sections will describe the
different modes of failure, as well as potential root causes of the fault.

2.1 Resistive Leak Fault on HVDC Bus
Figure 1 illustrates a typical construction of the HV power system on a HEV. As
can be seen, there are several locations on the HVDC power path that can leak to chassis.

2.1.1 Wire Chafing
Technological advancement of vehicles for increased safety, fuel efficiency, on
board diagnostics & prognostics, and artificial intelligence has resulted in significant
electrification of the vehicle system. This electrification has been predominantly low
voltage in nature, but with the advent of hybrid electric vehicles, high voltage power
systems including the power lines and connectors have been introduced to the
environment. Although these HV lines are usually very well insulated and routed,
proximity to the vehicle chassis and potential abrasive surfaces is inevitable due to the
space and weight constraints on vehicles. Mechanical vibration of these cables causes

4

gradual wear and tear due to friction with surfaces in close proximity, eventually leading
to insulation failure and a subsequent ground fault. Many industry reports, such as the
one created by the NASA Research Center for Wiring Fault Detection, endorse that wire
chafing is the root cause of ~30% of all wiring faults that occur in aircraft [1].

Figure 1: Typical HV Power System on a Hybrid Electric Vehicle

5

2.1.2 Aging of Components
Due to the nature of plastic composite material used as insulation for vehicle
wiring, embrittlement and eventual cracking of wire insulation is a common
phenomenon. Cracked insulation reduces electrical isolation between a high voltage
conductor and the vehicle chassis, and hence can increase leakage current. Studies have
shown that electrical stress conditions, chemical structures, and operating environments
(both mechanical and thermal) lead to the aging and subsequent dielectric breakdown of
polymeric insulation [2].

2.1.3 Contaminant Intrusion
Sealing of electronic components in medium and heavy duty commercial hybrid
vehicles is a critical component towards system reliability, and is most often a challenge
because of the harsh environments that these vehicles are exposed to. The following
conditions could lead to the intrusion of contaminants into the HV system, eventually
leading to current leak paths to the chassis:
•

Constant heating and cooling of components resulting in condensation, and
subsequent moisture intrusion

•

Heavy rains or snow accumulation resulting in water entering improperly sealed
components, or those with porous metal enclosures

•

Broken, defective, or improperly sealed high voltage power connectors resulting
in water intrusion

6

•

Infiltration of salt, dirt, and other debris into high voltage components or battery
system

•

Leakage of electrolyte from high voltage battery cells to the case of the battery
which is grounded to the vehicle’s chassis [3] [4]

2.1.4 Manufacturing Defect
As power electronics systems become more complex in design due to the
integration of several power conversion modules into one chassis, manufacturability
becomes a challenge. This coupled with the fact that market demand for these
components is still relatively very low, most of the production is done manually. This
increases opportunity for build error, and eventual defects in parts which, depending on
the nature of the defect, might not surface until the component has been installed in the
vehicle and operated for several duty cycles. Some good examples of defects that could
cause ground faults are:

•

Wrong or defective components being installed in power system components. E.g.
a shorted Y-capacitor installed in an inverter

•

Foreign conductive material intrusion into the component during manufacturing.
E.g. extra screws, wires, components etc.

2.1.5 Maintenance worker incident
A majority of the tools (screw drivers, wrenches etc.) used during maintenance
and service are conductive in nature, and tight spaces in power systems enclosures could
cause the service personnel to short a high voltage cable or bus bar to the chassis of the

7

vehicle. If a pre-existing ground fault is present in the system (from either positive or
negative conductor to chassis), and the service personnel creates a fault between the other
conductor and chassis, an arc condition could exist and potentially harm the individual, in
addition to damaging the tool.

2.2 Alien Component Introduced to System
High Voltage battery power on hybrid electric vehicles although traditionally was
used to drive a traction motor for propulsion, increased vehicle electrification has resulted
in the batteries being a source of power for several other components such as power
steering systems, air conditioning systems, 120V auxiliary power generators, HV battery
chargers etc. Currently, few industry standards exist that regulate the design of these
devices; which means that such commodity high voltage components could potentially be
used in a flexible power system, and cause a ground fault.

2.2.1 Excessive filtering to ground
Electromagnetic Interference (EMI) has become a great concern in power
electronics applications, especially in Hybrid Electric Vehicles, and the problem seems to
be increasing as the industry is moving rapidly towards higher power density components
for smaller package sizes. Although certain design procedures can be used to reduce
component sizes (such as higher switching frequencies), these same procedures worsen
the EMI phenomenon. In order to reduce EMI within a system, engineers use various
filtering techniques in a power device. The size of these filters is related to the desired
degree of attenuation of harmonics of the current noise [5].

8

Figure 2: Differential Mode and Common Mode filters [5]
Figure 2 illustrates a typical EMI filter in a power converter system, where CY and
LC are inductive and capacitive elements used to suppress Common Mode (CM) noise,
and LD and CX are used to suppress Differential Mode (DM) noise. CM noise occurs on
all power supply lines with respect to the reference ground plane and it is essentially
caused by insulation leakage, electromagnetic coupling and secondary effects due to
parasitic components. DM noise is always present between the two power supply lines
and it is mainly caused by pulsating currents and device turn-on and turn-off transients
[5].
Although CM noise filter elements reduce system EMI, they contribute
significantly to AC leakage current within the system. In a modular HEV where there
exist more than five power converter devices, each with its own EMI filter, the
compounding effect of the filter components could lead to leakage currents that have
hazardous effects.

9

2.2.2 Auxiliary Loads with Pre-existing Faults
Many power electronics devices, especially in hybrid vehicle applications, are
designed to have isolation between the High Voltage DC input stage and the output AC
or DC stage. This is most commonly achieved through galvanic isolation methods, such
as transformers. However, galvanic isolation is not a mandate in the industry, and there
are power electronics manufacturers that do not design isolation into inverter devices in
order to save on size and cost. A typical example of this is a non-isolated Auxiliary
Power Generator (APG) device that converts 300-400 VDC to 120VAC at 60Hz for
accessory usage in Utility truck HEVs or heavy duty line hauler truck HEVs[6]. The lack
of isolation causes the high voltage bus to be physically connected to any device that the
user plugs into the system, such as hand tool battery chargers, halogen lamps, fans, air
compressors, microwave ovens, air conditioners etc. The system therefore is vulnerable
to leakage current to chassis ground through these devices if a fault exists within them.

2.3 Leak Fault on the HVAC Bus
Depending on the topology of the power converter, isolation breakdown between
the output High Voltage AC lines to chassis could lead to excess flow of leakage current
to ground. The root causes of an AC side resistive fault are similar to those outlined in
Section 2.1. Furthermore, excessive filtering on the common mode AC bus could also
add to ground leak current. However, since there are power semiconductor devices
(IGBTs, MOSFETs etc.) in the current path between the AC load and the HVDC energy
source, the profile or footprint of this leakage current will be distinctly unique when

10

compared to DC side leakage current. Characterization of such a fault therefore has to be
done differently.

2.3.1 Motor winding insulation fault
Ground faults are more prevalent in motors than other power systems devices,
because of the violent manner and frequency with which they are started. The number
one contributor to motor winding insulation failure is thermal stress, which can be a
result of heat loss in the windings, high ambient temperature conditions, or friction of
insulation caused by severe vibration profiles [7]. Such a ground fault could cause
additional thermal stress due to the fault current, voltage stress to the inverter, higher
electrical noise emissions, increased motor torque ripple etc.
Ground current also flows through the leakage capacitance and inductance
intrinsic to the machine windings. When an induction or Permanent Magnet Synchronous
Machine is driven by a PWM inverter, as shown in Figure 3, the triangular carrier
frequency is often set beyond the audible frequency range (over 15 kHz) in order to
suppress acoustic noise. In this case, the output voltage potential of the inverter steps up
and down at a very high frequency according to the switching of the transistors or FET's
and the charging current flows into the leakage capacitance formed between the windings
and the iron core of the motor. This current flows through the capacitance to the ground
and then returns to the inverter along a miscellaneous path. The leakage current consists
of usually spike wise pulses which correspond to the switching instants of the transistors,
and the RMS value of the current is roughly proportional to the square root of the number

11

of the pulses or to the switching frequency. Since the core of the machine is grounded to
the chassis of the vehicle, this leakage current could be a shock hazard for personnel [8].

Figure 3: Equivalent circuit of traction inverter and induction motor in HEV [8]

Figure 4: Speed and Torque during phase to ground fault [7]

12

CHAPTER 3
3 Effects of Ground Faults
In the previous chapter, various failure modes and root causes related to ground
faults in HEVs were discussed. Any degree of reduction in the overall impedance of the
High Voltage power system with respect to the original system design could result in
various unwanted system behaviors or conditions.

3.1 Interference with other Circuits
Several circuits exist within subsystems of the HEV that are designed for device
control, data acquisition, prognostics, diagnostics, and protection from conducted and
radiated emissions. Some of these circuits use voltage and current measurements of the
high voltage bus with respect to chassis in order to operate effectively. A ground fault
between a high voltage conductor and vehicle chassis could result in the malfunction of
these electronic circuits, often resulting in unwanted shut down of the hybrid system,
increased vehicle downtime, and potential warranty costs to be incurred by the
manufacturer.

3.2 Potential Shock Hazards (AC and DC current)
By far the most critical reason ground fault protection is an absolute necessity in
power electronic systems is the potential risk a ground fault poses to the safety of people.
Traditional electrical components on a vehicle operate on 12VDC, which is too low a
voltage to cause harmful currents to flow through the human body. However, most hybrid

13

vehicles today run on voltages ranging from 300 V to 1000 V; shock currents at these
voltages could be lethal.
For this reason, all hybrid vehicles with high voltage energy sources have floating
grounds, which essentially means that the High Voltage bus is completely isolated from
the low voltage bus.
A floating system ensures two layers of protection from a shock hazard, since
both the positive and the negative conductors are isolated from the user. This ensures that
a dual point failure has to occur to cause a shock, and thus is mandated as a design
practice for all high voltage systems. In the case that a low impedance path exists
between a high voltage rail (positive or negative) and chassis as shown in Figure 7 on
Page 18, however, a single point failure is sufficient to cause a shock. It is therefore very
critical to have a system that monitors and reacts to a leakage condition within the hybrid
vehicle.
Many international organizations have released literature and standards on the
effects of electrical shock on human physiology, and have mandated certain safety design
practices to regulate the quality of products being released in the market.
Figure 5 below describes the various effects of increasing current (at 60Hz) on the
human physiology. As can be seen, non-hazardous leakage current is considered to be
any amount below the let-go current threshold. Hence, most ground fault circuit
interrupting devices are designed to open the circuit below this threshold.

14

Shock current @

Physiological Effect

60 Hz (1 – 3 sec)
0.2 mA – 7 mA

Threshold of Perception

8 mA – 90 mA

Let-go current

> 15 mA

Respiratory paralysis, fatigue, pain

> 70 mA – 5 A
>1A

Ventricular fibrillation
Sustained myocardial contraction, burns,
injury

Figure 5: Physiological effects of electricity [9]
In Figure 5, threshold or estimated mean values are given for each effect in a 70 kg
human for a 1 to 3 s exposure to 60 Hz current applied via copper wires grasped by the
hands [9].
UL 2231-2 – “Personnel Protection Systems for Electric Vehicle Supply
Circuits,” is a standard authored by the Underwriter’s laboratory and has a very robust
test methodology for determining the effectiveness of ground isolation within a system,
as well as that of circuit interrupting devices designed to protect personnel in the case of a
ground fault [10].
In summary, this standard introduces a distinct means to measure leakage current
through a human body model, and quantify the impact the shock current has on the
human physiology. By placing the measuring instrument circuit (Figure 6) as a load
across the terminals of a power system that the user could potentially get access to in a
fault condition, one can measure the Measurement Indication Unit (MIU) which
correlates to a normalized current parameter (across various frequencies). MIUs are

15

related to the physiological effects when electric current flows through the human body.
At low frequencies, the number of MIUs that is obtained by dividing the output voltage,
in mill volts, by 500 ohms, equals the current, in milliamperes, through the measuring
instrument. At high frequency, the meter indication of MIUs is less than the number of
milliamperes through the measuring instrument. For example, at 100 KHz, 0.5 MIU-RR
occurs when the actual current through the measuring instrument is 70 mA. At any
frequency, the acceptability of the leakage current can be determined by comparing the
number of MIU’s to the MIU limit, which is shown in Table 1.
This measurement methodology forms the basis for designing and testing ground
fault detection circuits in HEVs.

Figure 6: Equivalent Human Body Circuit Model [10]

16

Table 1: Required trip threshold for charger circuit interrupting devices [10]
Type of source for fault-current

Ground fault threshold – I
(rms mA) or (MIU)

60 Hz

5±1

DC

30

DC+AC

5 minimum

AC > 60 Hz

5 × FF but not greater than 70

Multiple frequencies

5 × FF but not greater than 70

FF – is the Frequency Factor from Figure 21.2 on Page 30 of [10]

3.3 Damage of Auxiliary Filter Circuits
Although isolating the high voltage bus from the chassis of the vehicle adds
several layers of safety from an electrical shock, it makes it difficult to shunt electrical
noise (generated by power electronics circuits) to the vehicle chassis. This noise can be
electromagnetically coupled into other subsystems in the vehicle, and cause interference
with their operation. It is therefore common practice to design EMI filters into power
electronics components to reduce the conducted and emitted noise footprint of the
component. These filters are usually made up of small y-capacitors or common mode
chokes (inductors). This addition of discrete components not only adds cost and
complexity to the device, but also leaves them susceptible to damage in the case of a high
voltage short circuit to the chassis.
Consider a common mode EMI filter as shown in Figure 7, where the filter
impedance is balanced between High Voltage positive and chassis, and High Voltage

17

negative and chassis. In an ideal state, the voltage potential across ‫ܥ‬ଵ and ‫ܥ‬ଶ would be

equal, i.e. half of the source DC voltage.

‫ܥ‬ଵ ൌ ‫ܥ‬ଶ ൌ 1ൗ2 ܸ‫ܥܦ‬
Y-capacitors are generally rated for the full DC voltage, with some additional
voltage for margin. This ensures longevity of capacitor life and protects against damage
during fault conditions. For example, in the case of a DC side ground fault as illustrated
in Figure 7 below, the Y-capacitor ‫ܥ‬ଶ would not see more than 400V, assuming that the

DC voltage source is rated for a maximum of 350V. However, in the case that the traction
motor of the HEV has a back EMF characteristic of above 600V at high speeds, there
could be potential overvoltage (and eventual damage) of the capacitors.

Figure 7: Effect of back EMF on EMI filter
In the case of an AC side ground fault, high frequency current ripple of significant
magnitude could result in excessive heating of the Y-capacitors, eventually causing
accelerated aging and damage [11].

18

CHAPTER 4
4 Survey and Analysis of Existing
Technologies
Although ground fault detection units and circuit interrupters for power systems
in residential and industrial applications have been available for many decades now,
distinct differences in vehicle applications have created the need to develop unique
diagnostic circuits for HEVs. There have been many inventions of such diagnostic
systems for vehicles in the past 5 years. This chapter will discuss in detail the operation
of these systems, and how they may be implemented in a flexible high voltage system.

4.1 Definition of Flexible HEV Architecture
As mentioned earlier in the introduction, the commercial hybrid electric vehicle
space varies significantly from that of the automotive HEV industry. Due to the plethora
of applications, and relatively low production quantities for these applications, many
companies are opting to design flexible HEV systems that provide the Original
Equipment Manufacturer (OEM) with a base propulsion system (a HV energy source,
inverter, and motor) that has provisions for connectivity to other auxiliary devices that
serve key functions in the vehicle. This allows OEMs to install components that

are

pertinent to the utility of the HEV, and also offers them the flexibility to source
components from a multitude of suppliers that may manufacture components that achieve
the same function but are designed differently. The ultimate goal is to develop a system
that is modular in terms of the devices that can plug into it and scalable in terms of power

19

throughput, while being adaptable to any base vehicle; all at a reasonable cost [9]. In
order to do this, the architecture of the base propulsion system should have a robust
electrical and mechanical interface, with in-built software to control and monitor
auxiliary devices. This software embedded in the system controller should also be
capable of detecting ground faults accurately (both DC and AC side), irrespective of the
components that are plugged into the system.
Figure 8 illustrates this type of architecture:

Figure 8: Flexible Hybrid Electric Vehicle architecture [12]
As vehicle electrification is growing at a rapid pace, there are many more
components that can be designed for and connected to a high voltage power net within a
HEV. This change in design philosophy also lends towards progress in the Electric
Vehicle (EV) industry, where an ICE (Internal Combustion Engine) and 12V battery

20

source are not needed, since all components are powered by a High Voltage battery
source. Some examples of devices that can be connected to the baseline drive system are
DC-DC converters, Auxiliary Power Generators (APG) for 120V power, Power Grid
Connect interfaces, Heating Ventilation and Air Conditioning (HVAC) systems, brakes,
power steering systems, and many other electrical accessories.
The HV Energy Management System would house the ground fault detection
technology, which would ideally monitor for faults in the entire system. The next section
discusses the various technologies that exist in the industry today, and evaluates whether
they would be a robust solution for a flexible architecture as described above.

4.2 Nissan Motor – Ground Fault Detector for Vehicle
4.2.1 Technology Overview
Inventors Tsuyoshi Morita and Shinsuke Nakazawa from Nissan Motor Co.
developed an adaptation of the popular signal injection method for detecting ground
faults in an electric vehicle (or HEV). This ground fault detector system interfaces with
the High Voltage Bus of the vehicle, as well as the low voltage system, via a controller
[13].
The system described in the invention is designed to perform three operations:
1. Detect the presence of a ground fault within the system
2. Identify the location of the ground fault within the system
3. Identify the cause of the ground fault

21

The method allows for perpetual monitoring of the HV bus for ground faults, even
during the operation of the vehicle. The following diagram may be used to illustrate the
invention:

Figure 9: Schematic overview of method 1 [13]
As can be seen from Figure 9, a square wave signal generator is used to output a

square pulse of amplitude ܸௌ (e.g. 5V) and frequency ݂ (e.g. 10Hz). The output of the
ଵ
generator is capacitively coupled to the high voltage positive (+) rail through capacitor

‫ܥ‬ଵ . Per the patent, the signal frequency of the square pulse should be significantly less

than the switching frequency (usually 20 kHz to 3MHz) of power electronics components
on the HV bus in order to prevent the high frequency noise from interfering with the leak
detection circuit’s ability to operate effectively.

22

The ground fault detector then measures the voltage (ܸ஽ா் ) at point A in the

circuit, to determine the attenuation in the signal based on common mode impedance in
the HV bus. This equivalent common mode impedance (ܼாொ_௉ ) between the high voltage

positive rail and chassis, and (ܼாொ_ே ) between the high voltage negative rail and chassis

is a combination of all the resistive, capacitive, and inductive elements in the common
mode bus. For simplicity, only resistive and capacitive elements will be considered in this
analysis.
As described in the previous section, consider an HEV system with a HV battery

source (ܸ஻஺்் ) and several fixed and variable power electronics components connected

to this source (Units 1, 2, 3, 4 etc). Each component has inherent impedance between the
high voltage rails and chassis ( ܼଵ_௉ , ܼଵ_ே , ܼଶ_௉ , ܼଶ_ே etc) where:

ܼଵ_௉ ൌ

൬

And

ܼଵ_ே ൌ
Therefore,

ܼாொ_௉ ൌ
And

൬

1

ܼଵ_௉

൬

1

1
൰ ൅ ݆൫2 ߨ݂‫ܥ‬ଵ_௉ ൯
ܴଵ_௉
1

ܴଵ_ே

൰൅൬

1

൰ ൅ ݆൫2 ߨ݂ ‫ܥ‬ଵ_ே ൯
1

ܼଶ_௉

23

1

൰൅൬

1

ܼଷ_௉

൰൅൬

1

ܼସ_௉

൰

ܼாொ_ே ൌ

൬

1

ܼଵ_ே

൰൅൬

1

ܼଶ_ே

1

൰൅൬

1

ܼଷ_ே

൰൅൬

1

ܼସ_ே

൰

There also potentially exists a resistive fault element ܴ௅_௉ or ܴ௅_ே between the

high voltage rails and chassis (ground), and ܴ௅_஺஼ between the non-isolated switched

AC line of an inverter device (e.g. single phase auxiliary inverter) and the chassis.

Ideally, ܼாொ_௉ and ܼாொ_ே are relatively large for DC and low frequency AC.

Hence, the attenuation in signal ܸௌ (measured as ܸ஽ா் ) could be relatively small in a

faultless system. The amplitude of ܸ஽ா் is as follows:

ܸ஽ா் ൌ ܸௌ ‫ڄ‬

ܼாொ_௉ െ ݆ܺ‫ܥ‬ଵ
ܼாொ_௉ െ ݆ܺ‫ܥ‬ଵ ൅ ܴଵ

The reactance of ‫ܥ‬ଵ is an order of magnitude smaller than ܴଵ at 10 Hz. Hence, in

the case that a resistive fault ܴ௅_௉ or ܴ௅_ே (e.g. 5

) exists, ܼாொ_௉ or ܼாொ_ே becomes

significantly small as well, resulting in large attenuation of the 5V signal ܸௌ . Based on
this principle, an appropriate threshold impedance ்ܼு may be chosen as a lower limit

for the common mode impedance ܼாொ_௉ or ܼாொ_ே , and the corresponding threshold

voltage ்ܸு can be derived to be used as a metric for diagnosing ground faults.

This summarizes the ground fault detection methodology of the invention. The
authors also discuss a simple means for detecting the location of the fault, by first
detecting its presence in the system, and then isolating the fault to individual components
by controlling switch circuits 4A, 4B, 4C, and 4D independently at different times.

24

Finally, the invention also discloses a method to determine the nature of the fault

within the system by varying the oscillating frequency of signal ܸௌ . The controller

generating the square pulse first changes the frequency to ݂ଶ , which is lower (e.g. 5Hz)

than the original frequency ݂ . In the case of a resistive fault, the impedance ܼாொ_௉ െ
ଵ

݆ܺ‫ܥ‬ଵ or ܼாொ_ே െ ݆ܺ‫ܥ‬ଵ changes by the same factor as the frequency, i.e.:
ܼாொሺ௙ଶሻ െ ݆ܺ‫ܥ‬ଵሺ௙ଶሻ ൌ ሺܼாொሺ௙ଵሻ െ ݆ܺ‫ܥ‬ଵሺ௙ଵሻ ሻ ‫ڄ‬

݂ଶ
݂ଵ

However, in the case of a capacitive fault or a fault of dielectric nature, the

impedance ܼாொ െ ݆ܺ‫ܥ‬ଵ at frequency ݂ଶ will be higher than that of a resistive fault at

frequency ݂ଶ , since the reactance of capacitive elements is nonlinear with respect to time,
as shown in Figure 10 below:

Figure 10: Change in impedance for capacitive vs. resistive ground faults [13]

25

At the decreased frequency, the threshold amplitude ்ܸு (or fault judgment

value) for fault detection is set such that it is higher than the signal amplitude for a
Hence, based on the attenuation factor of the detection voltage ܸ஽ா் after the frequency
resistive fault, but lower than that for a capacitive fault (see Figure 11 for illustration).

is decreased, the diagnosis could determine whether the existing ground fault is resistive
or capacitive in nature.

Figure 11: Response of circuit after decrease in frequency [13]
Similarly, the original frequency can be increased to ݂ଷ (e.g. 20 Hz) to identify

the nature of the fault. At the increased frequency, the threshold amplitude ்ܸு for fault
26

detection is set such that it is lower than the signal amplitude for a resistive fault, but
higher than that for a capacitive fault (see Figure 12 for illustration).

Figure 12: Response of circuit after increase in frequency [13]

4.2.2 Simulation and Analysis
The method proposed by the inventors is a technique that has been in use in other
industries, such as Uninterruptible Power Supplies, for many years prior to the invention
[3]. In order to analyze the viability and effectiveness of the method, this paper will test

27

the proposed ground fault detection circuit in seven different test scenarios that may exist
within a HEV system.
Figure 13 on Page 31 is a circuit model designed using the Allegro Design Entry
CIS application to replicate the above ground fault detection system, along with common
mode impedances of power electronics components in the system.
As can be seen from the circuit, values for ܴଵ (54 k ) and ‫ܥ‬ଵ (2.65uF) have been

chosen arbitrarily based on the following criteria:

݂ଵ ൌ 10‫ݖܪ‬
ܸௌ ൌ 5ܸ

்ܸு ൌ 3.45ܸ

ܸ஻஺்் ൌ 600ܸ

‫்ܫ‬ு ሺthreshold leakage current to trip ground faultሻ ൌ 5݉‫ܣ‬

்ܼு ሺthreshold impedance to trip ground faultሻ ൌ
ܴଵ ൌ 54݇Ω

ܸ஻஺்்
ൌ 120݇Ω
‫்ܫ‬ு

We know that attenuation of ܸௌ will be 31% at the threshold impedance, i.e.:
ଶ
ଶ
ඥܺܿଵ ൅ ்ܼு

ଶ
ଶ
ඥܺܿଵ ൅ ்ܼு ൅ ܴଵ

ܺܿଵ ൌ

ൌ

1
2 ߨ ݂ଵ ‫ܥ‬ଵ

28

3.45ܸ
5ܸ

Using the above equations and known values, we solve for ܺ஼ଵ = 6005

Therefore, ‫ܥ‬ଵ ൌ 2.65‫ܨݑ‬

ܴଶ and ‫ܥ‬ଶ in the circuit are an added filter that would typically serve to eliminate

high frequency noise that is superimposed onto ܸ஽ா் because of semiconductor

switching (this filter is not specified in the invention disclosure).

The cutoff frequency (݂஼ ) of the added filter is ~16Hz, so that the 10Hz

component of ܸௌ is not attenuated significantly.

݂஼ ൌ

1
1
ൌ
ൌ 15.9 ‫ݖܪ‬
ܴଶ ‫ܥ ڄ ߨ2 ڄ‬ଶ 100 ‫01 ڄ‬ଷ ‫ି01 ڄ 1.0 ڄ ߨ2 ڄ‬ଷ

We can calculate the new threshold voltage for ground fault detection using the known
time constant ܴଶ ‫ܥ‬ଶ of the filter:

்ܸு_ோௐ ൌ ்ܸு ‫ڄ‬

ܺ஼ଶ ൌ ൬

்ܸு_ோௐ ൌ 3.45 ൈ

ܺ஼ଶ

ଶ
ଶ
ඥܺ஼ଶ ൅ ܴଶ

1
൰ ൌ 159.2݇Ω
2 ߨ10 ‫ܨݑ1.0 ڄ‬
159.2 ൈ 10ଷ

ඥሺ159.2 ൈ 10ଷ ሻଶ ൅ ሺ100 ൈ 10ଷ ሻଶ

ൌ 2.92ܸ

Hence, if ܸ஽ூை஽ா is being used as the measurement signal, the threshold for

ground fault detection would be 2.92V.

29

A diode ‫ܦ‬ଵ with a breakdown voltage of 5V has also been added to the leak

detection circuit to prevent high voltage transients from damaging the controller input.

A High Voltage battery source, traction inverter, auxiliary single phase inverter, DCDC
converter, and one auxiliary load have been modeled in the simulation circuit. Switches
S1 through S10 are used to connect or disconnect the various power electronics
components to or from the HV battery source.
TEST CASE 1
For the first test case, the ground fault detection circuit is connected only to the
HV battery and the traction inverter. Also, there are no low impedance paths between the
high voltage rails and chassis ground. This test case is to evaluate the ideal response of
the ground fault detection system during a no-fault condition, and verify that a false fault
would not be set.
Hence, all switches in the system are open, and the only impedance to ground in

addition to the capacitor ‫ܥ‬ଵ is the common mode impedance of the traction inverter:

ܼଵ_௉

ଶ
ଶ
ܴଵ_௉ ‫ܥ ڄ‬ଵ_௉
ܴଵ_௉ ‫ܥ ڄ‬ଵ_௉
ൌ ଶ
ଶ െ݆ ‫ڄ‬ቆ ଶ
ଶ ቇ
ܴଵ_௉ ൅ ‫ܥ‬ଵ_௉
ܴଵ_௉ ൅ ‫ܥ‬ଵ_௉

Based on Figure 14, as expected, there is a small amount of attenuation in the

signal, and ܸ஽ா்ሺ௣௣ሻ ൌ 4.39ܸ . Since the threshold voltage is 3.45 V, no fault is

triggered.

30

Figure 13: High voltage power system circuit model
31

Figure 14: Simulation results for test case 1 of method 1
(For interpretation of the references to color in this and all other figures, the reader is
referred to the electronic version of this thesis)
TEST CASE 2
For the second test case, switches S3, S4, and S5 will be closed in order to
connect the Auxiliary Single Phase Inverter. Since the inverter is not galvanically
isolated, the input HVDC bus and output AC bus are physically connected through the
switching IGBTs. Hence, high frequency common mode noise due to IGBT switching, as
well as 120Hz ripple voltage (two times the fundamental output frequency of inverter)
will be superimposed onto the signal ܸ஽ா் . This test case is simulated to understand the

susceptibility of the circuit to ripple voltage and switching noise. The results are shown in
Figure 15.

32

Figure 15: Simulation results for test case 2 of method 1

ܸ஽ா்ሺ௉௉ሻ ൌ 16.9ܸ

ܸ஽ூை஽ாሺ௉௉ሻ ൌ 4.92ܸ

There are two components that contribute to the overall 120Hz waveform being

superimposed onto ܸ஽ா் :

1) The DCR (Direct Current Resistance) of the HV Battery system as well as
inherent series impedance of the transmission path from the battery to the inverter
causes a voltage ripple on the HVDC link when the single phase inverter load is
drawing 120Hz current.
2) In a non-isolated inverter system, the ratio of DC side impedance between high
voltage and ground to the overall impedance of the system (AC and DC side) will
determine the amplitude of the AC voltage that will be superimposed onto the DC
bus. Figure 17 describes this.

33

Figure 16: DC ripple current in a non-insulated single phase inverter system

Figure 17: Equivalent circuit of AC and DC common mode bus

ܸோூ௉௉௅ா ൌ ܸ஺஼ଵ ‫ ڄ‬൬

ܼ஽஼
ܼ஽஼
൰ ൅ ܸ஺஼ଵ ‫ ڄ‬൬
൰
ܼ஽஼ ൅ ܼ஺஼ଵ
ܼ஽஼ ൅ ܼ஺஼ଶ

ܸோூ௉௉௅ா is the ripple voltage on the high voltage bus

ܸ஺஼ଵ and ܸ஺஼ଶ are common mode output voltages of the inverter

ܼ஽஼ is the common mode impedance between HV + and chassis, and HV – and chassis

34

ܼ஺஼ is the common mode impedance between inverter output line 1 and chassis, and line
2 and chassis

The red line in Figure 15 represents ܸ஽ா் , and has a peak-to-peak amplitude of

16.9V. For most microcontrollers, the analog or digital inputs are limited to 5V; hence,
this would be too high a voltage to accurately measure. This is where the second filter

(ܴଶ , ‫ܥ‬ଶ ) and the zener diode (‫ܦ‬ଵ ) come into play. ܸ஽ூை஽ா is shown in blue in Figure 15,
and is representative of the filtered version of ܸ஽ா் .

The peak-to-peak value of ܸ஽ூை஽ா is 4.92V, which would clearly not trip a leak

fault (the threshold for ܸ஽ூை஽ா being 2.92V). This is an important characteristic to note,
since the 120Hz ripple on the signal adds to the complexity of signal processing of the

measured signal, since there is poor fidelity in the raw waveform. Depending on the
architecture of the system, this could be a significant issue. Although the most effective
manner for processing ܸ஽ா் would be to sample at a very high frequency (e.g. 1kHz) and

then calculate a moving average of the data points to calculate the RMS voltage of the
signal, the new RMS value might still misrepresent the true condition of the system. On
the other hand, if the peak-to-peak voltage was to be used for ground fault analysis, the
data would be highly unreliable due to the added noise on the signal (as illustrated in this
test case).
TEST CASE 3
In this test case, the response of the ground fault detection circuit during an actual
fault on the DC side is analyzed. Switch S1 between High Voltage Positive and chassis is

35

closed at time ‫ ݐ‬ൌ 0 ‫ .ܿ݁ݏ‬All other switches in the system are open, and the leakage

resistance ܴ௅_௉ is set to 120k . Hence, the only common impedance in the HV positive

to chassis ground system would be that of the leakage resistance ܴ௅_௉ in parallel with

‫ܥ‬ଵ_௉ and ܴଵ_௉ (in the traction inverter). This test case is to validate that the circuit can
successfully detect a fault in a true leakage condition. The result of the simulation is
shown in Figure 18 below.

Figure 18: Simulation results for test case 3 of method 1

ܸ஽ா்ሺ௉௉ሻ ൌ 3.38ܸ

ܸ஽ூை஽ாሺ௉௉ሻ ൌ 3.05ܸ

36

As expected, ܸ஽ா் is less than the threshold voltage ்ܸு ൌ 3.45ܸ ; this value

would trigger a ground fault and allow the system controller to take necessary actions.
The ground fault detection method described works well for purely resistive faults.
TEST CASE 4
In this test case, not only is SW1 closed to create a leakage resistance path to
ground from the HV positive rail, but also switches SW3, SW4, and SW5 in order to
connect the Single Phase Inverter. In test case 2, it was demonstrated that the diagnosis of
a ground fault could become complex depending on the robustness of the Digital Signal
Processing (DSP) within the microcontroller. It is important to understand if a true
ground fault can be detected in the case of ripple voltage being superimposed on the
measured signal ܸ஽ா் . The results are shown in Figure 19 on Page 38.

37

Figure 19: Simulation results for test case 4 of method 1

ܸ஽ா்ሺ௉௉ሻ ൌ 15.55ܸ

ܸ஽ூை஽ாሺ௉௉ሻ ൌ 4.16ܸ

ܸ஽ூை஽ாሺோெௌሻ ൌ 2.62ܸ

It is apparent that the amplitude of the filtered signal ܸ஽ூை஽ா (peak to peak) is

less than it was in test case 2; however, if the peak to peak voltage was used as a

parameter for ground fault diagnosis, the system proposed in this invention would not be
able to correctly detect that a ground fault exists, since ܸ஽ூை஽ா ൌ 4.16ܸ ൐

்ܸு_ோௐ ൌ 2.92ܸ .

If the RMS voltage was to be used, the system would potentially be able to detect the
fault. As iterated earlier, the effectiveness of the proposed method is reduced in a system
with low frequency ripple on the common mode bus.

38

TEST CASE 5
In this test case, the impact of excessive DC side Y-capacitance on the ground
fault detection circuit is analyzed by closing SW3, SW4, SW5, SW7, SW8, SW9, and
SW10. Although the single phase inverter is now connected, the device is assumed to not
be converting power (hence, no 120Hz ripple).
Although Y-capacitance can be a leak path for AC current to flow from the HV
rails to chassis ground, very little AC voltage potential exists on the DC side common
mode bus. Therefore, Y-capacitance in the system does not contribute significantly to AC
leak currents, and is an open circuit for DC leak currents. Hence, it should not be
considered when diagnosing DC ground faults.
One might argue that during single phase inverter operation, low frequency ripple
voltage (e.g.120 Hz) on the common mode bus may reach significant amplitudes that
could result in unacceptably high AC leak currents flowing through the Y-capacitors to
chassis, which could be a shock hazard. Although this is true, there are several design
methods that can be implemented to reduce this AC voltage. Moreover, the scope of this
invention is to detect resistive faults to ground, and not capacitive faults.
The Single phase inverter, DC-DC Converter, and Auxiliary Load in the test
circuit each add 50nF of Y-capacitance to the system. Although 50nF is an arbitrary
value, it is not uncommon for manufacturers of High Voltage DC systems to sometimes
exceed the maximum allowable Y-capacitance design value for 60 Hz, 120Vrms AC
systems (where 60Hz leak current is considered to be worst case as described in Chapter
3). This maximum value may be calculated as follows [14]:

39

‫ܥ‬௒_ெ஺௑ ൏ ൬

‫ܥ‬௒_ெ஺௑ ൏ ቆ

‫ܫ‬஼ெ_ெ஺௑
൰
2 ߨܸ஺஼ ݂

3.5 ‫ି01 ڄ‬ଷ

2 ߨ ‫ ڄ‬൫120 √2൯ ‫06 ڄ‬

ቇ ൌ 54.7݊‫ܨ‬

Figure 20: Simulation results for test case 5 of method 1

ܸ஽ா்ሺ௉௉ሻ ൌ 3.58ܸ

ܸ஽ூை஽ாሺ௉௉ሻ ൌ 12.99ܸ

It is evident from the simulation results in Figure 20 that the added capacitance on
the common mode bus alter the filter characteristics of the ground fault detection system,
hence attenuating the source signal ܸௌ significantly, resulting in ܸ஽ா்ሺ௣௣ሻ being very

close to the threshold voltage ்ܸு of 3.45V. Given this response, a resistive leak path

(between HV rail and chassis) larger than the threshold resistance 120k

could

potentially trigger a ground fault. Depending on the system’s reaction to a fault, a false

40

ground fault diagnosis such as this could result in unwanted shut down of the hybrid
system, increased vehicle downtime, and potential warranty costs to be incurred by the
manufacturer.
TEST CASE 6
This test case evaluates the response of the ground fault detection system to an
AC side leak fault. To achieve this, Switch SW6 in the test circuit is closed to inject a
120k

resistive path between the Line 1 output of the Single Phase Inverter and the

chassis ground. The simulation results are shown in Figure 21.
Ideally, the ground fault detection system should be able to distinguish between a
fault on the DC side and AC side of the HEV power system. In the case of the cited

invention, significant ripple voltage on the detected signal ܸ஽ா் increases the complexity

of accurately diagnosing a leak fault, since ܸ஽ூை஽ாሺ௣௣ሻ is much higher than the threshold
voltage ்ܸு_ோௐ of 2.92V. Averaging methods implemented in system software might
not be a viable option either, due to the magnitude of ripple.

41

Figure 21: Simulation results for test case 6(a) of method 1

ܸ஽ா்ሺ௉௉ሻ ൌ 18.6ܸ

ܸ஽ூை஽ாሺ௉௉ሻ ൌ 4.95ܸ

In the case of an AC leak fault of very low resistance (e.g. 1 ), the measured

signals ܸ஽ா் and ܸ஽ூை஽ா would be unreliable diagnostic inputs since the full magnitude
of AC voltage will be superimposed onto the common mode bus, as calculated below:

ܸோூ௉௉௅ா ൌ ܸ஺஼ଵ ‫ ڄ‬൬

Where ܼ஽஼ ൐൐ 1

ܼ஽஼
ܼ஽஼
൰ ൌ 120 ‫ ڄ 2√ ڄ‬൬
൰ ൎ 120 √2 ൎ 170ܸ
ܼ஽஼ ൅ ܼ஺஼
ܼ஽஼ ൅ 1

The results of the simulation are shown in Figure 22.

42

Figure 22: Simulation results for test case 6(b) of method 1
TEST CASE 7
Although there are hybrid vehicle or electric vehicle architectures that operate on
a regulated High Voltage bus, many HEV platforms use a variable High Voltage battery
source to function. As a result, most power electronic components, such as a Voltage
Source Inverter (VSI) with input bulk capacitance, are designed to operate with an input
DC source that has a dynamic voltage profile [15]. The steady state voltage of a high
energy/ high power battery typically used in a HEV (e.g. 500V lithium ion battery pack)
gradually changes based on the state of charge of the pack. However, due to high power
pulses in the drive cycle of the vehicle, there are several instances of voltage excursions
in the High Voltage bus. This is a direct result of the voltage drop across the Direct
Current Resistances (DCR) of cells in the high voltage battery pack, during high current

43

pulses [16]. In a typical commercial vehicle application, there could exist current pulses
nearing 100A for a period of 5-10 seconds. Assuming an arbitrary DCR of 5݉Ω per cell

in a Li-ion battery pack with 150 cells in series, this would mean voltage excursions of
∆ܸ ൌ ‫ ܴܥܦ ڄ ܫ‬ൌ 100‫݉500.0 ڄ ܣ‬Ω ‫ 051 ڄ‬ൌ 75ܸ for 5-10 seconds [17].

This test case simulates a power discharge cycle of the High Voltage DC battery
by superimposing a random negative voltage waveform on the DC bus, as shown in
Figure 23 below, to determine the impact on the measured signal ܸ஽ா் .

Figure 23: Variation in the HV DC bus voltage during discharge cycle
These transients on the differential HVDC bus are also observed on the common
mode bus by virtue of the impedances between the HVDC positive and negative rails
and chassis ground. The frequencies of the transients are low enough to pass through

unfiltered to the leak detection circuit, resulting in erratic behavior of the signal ܸ஽ா் , as
44

shown in Figure 24 below. Once again, there is very low fidelity in the signal response
data, and diagnostics for ground faults become highly unreliable.

Figure 24: Simulation results for test case 7 of method 1
In conclusion, the method described in this section for ground fault detection is
fairly reliable in systems without large magnitudes of Y-capacitance, and those that have
isolated power electronics components. In general, however, it is not a suitable method
for a flexible hybrid power system as described in Chapter 4.1 for reasons outlined
throughout the analysis section.

45

4.3 Maxim Integrated Products – Fault Detecting
Method for Detecting Leak Paths between Power
Sources and Chassis
4.3.1 Technology Overview
Inventors Mark Plagens and Brian Fritz from Maxim Integrated Products, Texas
developed a method to detect a resistive ground fault condition between a DC power
system and the chassis of an EV or HEV. The aforementioned invention was developed
to address the need for a reliable detection methodology that would not be susceptible to
large amounts of noise created by parasitic or inherent capacitance in the common mode
HV bus, or a failed component within the detection circuit itself [18].
The following Figure 25 graphically illustrates the invention:

Figure 25: Schematic overview of method 2 [18]

46

The system described above is designed to detect the total parasitic resistance

ܴ௅௄ that exists between the High Voltage bus and chassis ground. A 1 resistor ܴௌଵ is
st

connected on one end to the positive node of the High Voltage system, and to a switch S1
on the other. The switch S1 when closed allows current to flow from the positive high

voltage rail to the chassis. This current ‫ܫ‬ଵ is measured by a current sensor device and then
sent to a microcontroller for processing. A 2

nd

resistor ܴௌଶ is connected to the High

Voltage negative rail on one end and to a switch SW2 on the other. Current ‫ܫ‬ଶ flows from

the negative rail of the bus to the chassis ground through SW2 when it is closed, and this
current is sensed and processed by the microcontroller in the system.
There also exists inherent capacitance within the system that could be a result of
close proximity of high voltage conductors to the chassis of the vehicle. In addition, there
are many power electronics components with Y-capacitors for EMI filtering, which could
add to the inherent common mode capacitance in the system. The fault detection system
functions as described below:
Switch SW1 is first closed at time ܶ଴ , causing current ‫ܫ‬ଵ to flow through the

resistor ܴௌଵ and switch SW1 until the inherent capacitance ‫ܥ‬ே (e.g. 1nF) is charged. As

‫ܥ‬ே is charged, the value of ‫ܫ‬ଵ declines, ideally approaching zero in a no fault condition
due to the large value of the ܴௌଵ (e.g. 1 M ) and the parasitic leakage resistance in the

system. At time ܶଶ , switch SW1 is opened and switch SW2 is closed. As a result,
inherent capacitance ‫ܥ‬ே discharges through SW2 and 2
47

nd

resistor ܴௌଶ (e.g. 1 M ), and

this current ‫ܫ‬ଶ eventually approaches zero. This behavior is illustrated in Figure 26, and

is representative of an ideal system with very large parasitic resistance.

In the case of a true leak fault, when SW1 is closed, current ‫ܫ‬ଵ does not decline

because of leakage current flowing through parasitic resistance (e.g. ‫ܫ‬௅௄ ൌ

10݉‫ ܣ‬through the parasitic resistance ܴ௅௄ ). At time ܶଶ , when SW1 is opened and SW2

closed, current ‫ܫ‬ଶ flows through the resistor ܴௌଶ but again does not decline due to the

leakage current across parasitic resistance ܴ௅௄ . The footprint of ‫ܫ‬ଵ and ‫ܫ‬ଶ in this case is
shown in Figure 27.

In the case where the parasitic resistance ܴ௅௄ is not very low, but is approaching

the threshold of 120k

(e.g. 300k ), current ‫ܫ‬ଵ declines rapidly but does not approach

zero due to the leakage across ܴ௅௄ . This behavior is illustrated in the Figure 28.

48

Figure 26: Response of I1 and I2 in a no fault condition [18]

Figure 27: Response of I1 and I2 in a true fault condition [18]

49

Figure 28: Response of I1 and I2 in a low impedance condition [18]
Similarly, in the case of a damaged component in the test circuit, such as an open

resistor ܴௌଵ , no current ‫ܫ‬ଵ will flow through the switch SW1 since the circuit is open.

Therefore, the signature of current waveforms ‫ܫ‬ଵ or ‫ܫ‬ଶ are sufficiently unique to diagnose

that the detection system has failed. This is one positive aspect of the described invention,
where every perturbation in the hybrid power system results in a response with a unique
signature that can be processed for reliable diagnostics. This can be achieved by
increasing the sampling rate of the current waveform, programming the microcontroller
to be able to identify its signature footprint, and then correlating the footprint to a certain
operating condition of the system. The patent for the invention provides more detail on
the mechanics of this function, which is not within the scope of this paper.

50

As the microcontroller acquires measurement data for ‫ܫ‬ଵ and ‫ܫ‬ଶ , It processes the

information to calculate the leakage current ‫ܫ‬௅௄ in the system, using the following
equations:

When SW1 is closed (and SW2 is open):

ܸ௉ െ ܸீ
ܴௌଵ ൅ ܴ௅௄

‫ܫ‬ଵ ൌ

Where ܸ௉ is the voltage potential at the HV positive node and ܸீ is chassis ground
potential

When SW2 is closed (and SW1 is open):

ܸே െ ܸீ
ܴௌଵ ൅ ܴ௅௄

‫ܫ‬ଶ ൌ

Where ܸே is the voltage potential at the HV negative

Combining the currents and setting ܴௌଵ ൌ ܴௌଶ :

ሺ‫ܫ‬ଵ െ ‫ܫ‬ଶ ሻ ൌ

ሺܸ௉ െ ܸீ ሻ ൅ ሺܸே ൅ ܸீ ሻ
ܸ஻஺்்
ൌ
ܴௌଵ ൅ ܴ௅௄
ܴௌଵ ൅ ܴ௅௄

Since ‫ܫ‬ଵ , ‫ܫ‬ଶ , ܸ஻஺்் , and ܴௌଵ are known values, ܴ௅௄ may be calculated as follows:

ܴ௅௄ ൌ ൬

ܸ஻஺்்
൰ െ ܴௌଵ
‫ܫ‬ଵ െ ‫ܫ‬ଶ

Leading to calculation of leakage current:

51

‫ܫ‬௅௄ ൌ

ܸ஻஺்்
ܴ௅௄

For accurate calculations of the leakage current, it is critical to use steady state

values of ‫ܫ‬ଵ and ‫ܫ‬ଶ (i.e. the set of data points in the tail end of the switch ON phase). The

derived value of ‫ܫ‬௅௄ may then be compared to a threshold leakage current value (e.g.
5mA) to diagnose a ground fault.

4.3.2 Simulation and Analysis
In order to perform a simulation of the fault detection system, the same HV
system circuit model shown in Figure 13 of Section 4.2.2 was used, with modifications
made to the leak detection circuit (as shown in Figure 29).

Figure 29: Method 2 leak detection circuit model
The following values were selected for the key parameters of the circuit (next page):

52

ܴௌଵ ൌ 1‫ܯ‬Ω
ܴௌଶ ൌ 1‫ܯ‬Ω

‫ܨ‬௦௪ଵ ൌ 1 ‫ ݖܪ‬ሺSwitching Frequency of SW1ሻ
‫ܦ‬௦௪ଵ ൌ 50% ሺON‐OFF Duty Cycle of SW1ሻ

‫ܨ‬௦௪ଶ ൌ 1 ‫ ݖܪ‬ሺSwitching Frequency of SW2ሻ
‫ܦ‬௦௪ଶ ൌ 50% ሺON‐OFF Duty Cycle of SW2ሻ

The switching frequencies of SW1 and SW2 are arbitrarily calculated based on

the known fixed inherent capacitance ‫ܥ‬ே of the system. Using a conservative design

approach, the ‫ܥ‬ே of the system would be the highest when all components are connected
to the High Voltage Bus. Assuming 100nF of Y-capacitance each (sum of capacitance
from HV positive to chassis and HV negative to chassis) for the Traction Inverter, DCDC Converter, Auxiliary Load, and Single Phase Inverter, this would amount to:

‫ܥ‬ே ൌ 100݊‫ 4 ڄ ܨ‬ൌ 400݊‫ܨ‬

The time constant of the circuit switched by SW1 and SW2 would then be (in a no fault
condition of ܴ௅௄ ൌ 1.5‫ܯ‬Ω):

ܴௌଵ ‫ܴ ڄ‬௅௄
1.5 ‫01 ڄ‬ଵଶ
߬ ൌ ܴாொ ‫ܥ‬ே ൌ ൬
൰ ‫ܥ ڄ‬ே ൌ ቆ
ቇ ‫ି01 ڄ 004 ڄ‬ଽ ൌ 240݉‫ܿ݁ݏ‬
଺
ܴௌଵ ൅ ܴ௅௄
2.5 ‫01 ڄ‬

Hence, a switch ON time of 500msec (50% Duty Cycle of 1Hz switching

frequency) would result in ‫ܥ‬ே charging to 1 െ ݁^ሺെ0.5/0.24ሻ ൎ 0.875 ൎ 87.5% of the

maximum voltage. This allows currents ‫ܫ‬ଵ and ‫ܫ‬ଶ to settle within 13% of their steady
53

state values, which means that under any condition, the system could have at most 13%
error in calculating the actual leakage current in the system (as described in section
4.3.1).
TEST CASE 1
As described in section 4.2.2, this test case is designed to evaluate the response of
the ground fault detection system in an ideal, no fault state. The conditions and
constraints for the test are the same as in section 4.2.2. The results are plotted in Figure
30 below.

Figure 30: Simulation results for test case 1 of method 2

As expected, the current ‫ܫ‬ଵ is highest at t = 0.5 sec, when the switch S1 is first

closed. As the capacitance ‫ܥ‬ே charges, the current gradually declines to a steady state

value at t = 1 sec. Subsequently, the reverse happens to ‫ܫ‬ଶ as the inherent capacitance ‫ܥ‬ே
54

initially discharges, and then eventually settles to a steady state value at t = 1.5 sec. The
leakage current can therefore be calculated as:

ܴ௅௄ ൌ

ܸ஻஺்்
600
െ ܴௌଵ ൌ ൬
൰ െ 1 ‫ ଺01 ڄ‬ൎ 3.5‫ܯ‬Ω
ି଺
ሺ65.872 ൅ 67.519ሻ ‫01 ڄ‬
‫ܫ‬ଵ െ ‫ܫ‬ଶ
‫ܫ‬௅௄ ൌ

ܸ஻஺்்
600
ൌ
ൌ 171.14ߤ‫ܣ‬
ܴ௅௄
3.5 ‫଺01 ڄ‬

Evidently, this is not greater than the leakage current threshold ‫்ܫ‬ு ൌ 5݉‫,ܣ‬

therefore a ground fault will not be set.
TEST CASE 2

The second test case is to evaluate the response of the circuit when a non-isolated
single phase inverter is connected to the HV battery. The conditions and constraints are
similar to those described in section 4.2.2. Unlike what was observed for the detection
methodology described in the previous section, there is little impact on the current
waveforms ‫ܫ‬ଵ and ‫ܫ‬ଶ that are used as inputs for diagnosis. This is illustrated Figure 32.

The leakage current in this case is 1.22mA, because of the resistors between high
voltage rails and chassis inside the inverter. Furthermore, the peak to peak amplitude of
the current ripple is 8.37uA (as shown in Figure 32), which is insignificant. The circuit
therefore is robust to low frequency voltage ripple on the common mode bus.

55

Figure 31: Simulation results for test case 2 of method 2

Figure 32: Zoomed simulation results for test case 2 of method 2

56

TEST CASE 3
This test case is to evaluate the response of the circuit in a true fault condition.
The conditions and constraints of this test case are similar to those in test Case 4 of
section 4.2.2. A 120k

leak fault is induced between the high voltage positive rail and

chassis, while the single phase inverter is connected to the system. The results are shown
in Figure 33.
The leakage current is calculated to be 5.70mA based on the simulation, which is
an accurate representation of the true leakage current.

Figure 33: Simulation results for test case 3 of method 2
TEST CASE 4
This test case is synonymous to test case 5 in section 4.2.2, where the response of
the leak detection circuit to added Y-capacitance the common mode bus is evaluated.

57

Figure 34: Simulation results for test case 4 of method 2

With a switching period of 500 msec, there is sufficient time for current ‫ܫ‬ଵ and ‫ܫ‬ଶ

to settle to a steady state current, in spite of 150nF of added capacitance ‫ܥ‬ே . The

calculated leakage current is 338.95uA, which once again is a fair representation of the
DC leak current from HV rail to ground.
TEST CASE 5

The conditions and constraints of this test case are similar to those of test case 6 in
section 4.2.2. The purpose of this simulation is to evaluate the response of the leak
detection system to an AC ground fault of 1
Figure 35 below.

58

(short circuit). The results are shown in the

Figure 35: Simulation results for test case 5 of method 2
There is a distinct signature to the footprint of the waveform in an AC leak fault
condition. As expected, a significant portion of the AC output of the single phase inverter
(@ 120Hz) is superimposed onto the common mode DC bus, causing AC leak currents to
flow through the capacitance CN. The amplitude of the AC leak current can be calculated
by measuring the peak and trough of the 120Hz component of the waveform, Therefore,
if the sampling frequency of the waveform was increased (e.g. 1000Hz), it would be
possible to identify the frequency and amplitude of the AC leak current, and diagnose an
AC leak fault within the system.

59

TEST CASE 6
The conditions and constraints of this test case are similar to those of test case 7 in
section 4.2.2. The purpose of this simulation is to evaluate the impact of voltage swings
in the DC bus on the current waveforms ‫ܫ‬ଵ and ‫ܫ‬ଶ . The results are shown in Figure 36.

Figure 36: Simulation results for test case 6 of method 2

As expected, the currents ‫ܫ‬ଵ and ‫ܫ‬ଶ are slightly skewed due to the swing in

voltage. However, this does not affect the calculation of leakage current ‫ܫ‬௅௄ because the
value of ܸ஻஺்் also changes dynamically with ‫ܫ‬ଵ and ‫ܫ‬ଶ . As a result, voltage swings in the

DC bus will not impact the reliability of the ground fault detection system.

60

4.4 Lear Corporation – Ground Fault Detection
System for Vehicles with High Voltage Power Net
4.4.1 Technology Overview
Inventors Miguel Angel Acena and Jordi Escoda from Lear Corporation,
Michigan developed a method to detect loss of electrical isolation in a vehicle that uses a
HV power net [19]. This is achieved by charging HV capacitors, then connecting them at
different times between the HV positive and negative rails and the vehicle chassis, and
measuring the remaining charge after a fixed period of time. The amount of energy lost in
the capacitors reflects the level of isolation of the floating HV power system to chassis
ground. If the discharge is low (shallow), the equivalent resistance to ground is high and
the isolation can be characterized as proper. If the discharge is above some nominal level,
the equivalent resistance is lower and isolation can be characterized as improper [19].
Once again, the primary purpose of the proposed methodology is to detect resistive leak
faults in the system, in both energized and non-energized states. There are minor
variations in the implementation of the fault detection system for a vehicle in the off-state
(when the HV batteries are disconnected) in comparison to a system for a vehicle in the
on-state (HV batteries are connected). However, the design principle behind both systems
is the same, and for the purpose of evaluating this method, we will analyze the fault
detection method designed for an energized system.
Figure 37 below shows an illustration of the Ground Fault Detection System
(GFDS).

61

Figure 37: Schematic overview of method 3 [19]

From Figure 37, ‫ܸܪ‬௉ and ‫ܸܪ‬ே are the High Voltage Positive and High Voltage

Negative power nets respectively. This typically represents the High Voltage bus of the

hybrid vehicle, which contains the HV battery pack and the power electronics
components of the system. ்ܸாௌ்_௉ and ்ܸாௌ்_௉ are two sources that are separate from
the HV battery, and are used to charge the capacitors ‫ܥ‬௉ and ‫ܥ‬ே through switches S1 and

S3 respectively. Switches S2 and S4 are used to connect each of ‫ܥ‬௉ and ‫ܥ‬ே to the HV

vehicle power lines (‫ܸܪ‬௉ and ‫ܸܪ‬ே ) for a fixed time. ܴீே஽_௉ and ܴீே஽_ே are the
inherent resistances of the high voltage system to chassis ground, and resistors ܴௌ_௉ and

ܴௌ_ே are intended to limit the discharge current from the capacitors ‫ܥ‬௉ and ‫ܥ‬ே in the
case of a very low value of ܴீே஽_௉ or ܴீே஽_ே (e.g. a short circuit) [19].

The aforementioned HV battery can be represented as an ideal voltage source plus

a low equivalent series resistor (ܴௌ_஻஺்் ). By applying Thevenin’s theorem to the

vehicle power net at ‫ܸܪ‬௉ and ‫ܸܪ‬ே with respect to chassis ground, the circuits in Figure

62

38 arise. Assuming ܴௌ_஻஺்் is negligible in comparison to ܴீே஽_ே and ܴீே஽_௉ , the
following equations may be derived:

ܴ௉_ாொ ൌ

ܴீே஽_ே ‫ீܴ ڄ‬ே஽_௉
ܴீே஽_ே ൅ ܴீே஽_௉

ܴே_ாொ ൌ

ܴீே஽_ே ‫ீܴ ڄ‬ே஽_௉
ܴீே஽_ே ൅ ܴீே஽_௉

ܸ௉_ாொ ൌ ܸ஻஺்் ‫ڄ‬

ܴீே஽_௉
ܴீே஽_ே ൅ ܴீே஽_௉

ܸே_ாொ ൌ ܸ஻஺்் ‫ڄ‬

ܴீே஽_ே
ܴீே஽_ே ൅ ܴீே஽_௉

Figure 38: Equivalent circuit of common mode bus [19]
The operation of the GFDS can be described by observing the timing of the four
switches S1, S2, S3, and S4. Assuming all switches are initially in the open state, switch
S1 is closed at time ‫ݐ‬ଵ to allow ‫ܥ‬௉ to be charged up to ்ܸாௌ்_௉ . Before ‫ݐ‬ଶ , the capacitor

is completely charged (i.e.ܸ஼௉ ൌ ்ܸாௌ்_௉ ) and then S1 opens. At ‫ݐ‬ଷ , S2 is closed and

kept in that state up to ‫ݐ‬ସ . During this interval (‫ݐ‬ௌௐଶ ൌ ‫ݐ‬ସ െ ‫ݐ‬ଷ ) the capacitor ‫ܥ‬௉ is
63

discharged through a resistance that equals the sum of ܴௌ_௉ and ܴ௉_ாொ . After the

discharging process, at time ‫ݐ‬஼௉ , the resulting voltage across the capacitor ‫ܥ‬௉ (i.e. ܸ஼௉ )

can be approximated by the following equation:

ܸ஼௉ሺ௧ୀ௧಴ುሻ ൌ ൫்ܸாௌ்_௉ – ܸ௉_ாொ ൯ ‫݁ ڄ‬

ି

௧ௌௐଶ
஼ು ‫ڄ‬൫ோೄ_ು ାோಸಿವ_ಶೂ ൯

൅ ܸ௉_ாொ

This voltage ܸ஼௉ (at time ‫ݐ‬஼௉ ) reflects the level of isolation that exists between

‫ܸܪ‬௉ and chassis ground (i.e. ܴீே஽_ாொ ) since it is related to ܴீே஽_ாொ as shown in the

equation above.

A similar process can be followed using the switches S3 and S4 in order to arrive

at the following equation for the voltage across the capacitor ‫ܥ‬ே (i.e. ܸ஼ே ) after it goes
through the discharging process:

ܸ஼ேሺ௧ୀ௧಴ಿ ሻ ൌ ൫்ܸாௌ்_ே – ܸே_ாொ ൯ ‫݁ ڄ‬

ି

௧ௌௐସ
஼ಿ ‫ڄ‬൫ோೄ_ಿ ାோಸಿವ_ಶೂ ൯

൅ ܸே_ாொ

From the equation above, it is evident that both ்ܸாௌ்_௉ and ்ܸாௌ்_ே must be

greater than ܸ஻஺்் . This guarantees that the capacitors ‫ܥ‬௉ and ‫ܥ‬ே are charged up to a

voltage higher than ܸ஻஺்் , ensuring that there is charge available for the HV power net

for any value of ܴீே஽_ாொ . This can be achieved in several ways, and the patent discloses
one method which uses a flyback converter to step up the vehicle’s 12V battery voltage
to the desired ்ܸாௌ்_௉ and ்ܸாௌ்_ே .

64

Figure 39 graphically illustrates the sequencing of the switches. The measured values of

ܸ஼௉ at time ‫ݐ‬஼௉ and ܸ஼ே at time ‫ݐ‬஼ே can then be used to calculate ܴீே஽_ாொ , and

subsequently compare the result to a threshold value ்ܴு (e.g. 120k ) to determine
whether a fault condition exists or not.

Figure 39: Sequence of switch operation for method 3 [19]

4.4.2 Simulation & Analysis
In order to perform a simulation of the GFDS, the same HV system circuit model
shown in Figure 13 of Section 4.2.2 was used, with modifications made to the leak
detection circuit (as shown in Figure 40).

65

Figure 40: Leak detection circuit model for method 3

Assuming that the maximum differential voltage of the HV bus (ܸ஻஺்் ) is 600V,

்ܸாௌ்_௉ and ்ܸாௌ்_ே have been arbitrarily chosen to be 800V, which is higher than
ܸ஻஺்் . The choice of parameters ‫ܥ‬௉ , ‫ܥ‬ே , ܴௌே , ܴௌே , ܴௌଷ , and ܴௌସ is dependant on the

switching times ‫ݐ‬ଵ through ‫ . ଼ݐ‬Time ‫ݐ‬ௌௐଵ ൌ ‫ݐ‬ଶ െ ‫ݐ‬ଵ is the length of time that switch

SW1 is ON and ‫ܥ‬௉ is charging, and ‫ݐ‬ௌௐଶ ൌ ‫ݐ‬ସ െ ‫ݐ‬ଷ is the length of time that switch
SW2 is ON and ‫ܥ‬௉ is discharging. Considering 120k

to be the threshold for low ground

resistance (ܴீே஽_ாொ ), the level of discharge of capacitor ‫ܥ‬௉ to be 90% of its original
voltage, and ‫ݐ‬ௌௐଶ to be 245 msec (leaving 5 msec as dead time between ‫ݐ‬ௌௐଶ and

‫ݐ‬ௌௐଷ ), the value of ‫ܥ‬௉ can be calculated as follows:

ܸ஼௉ሺ௧ୀ௧಴ುሻ ൌ ்ܸாௌ்_௉ ‫݁ ڄ‬
66

ି

௧ௌௐଶ
஼ು ‫ڄ‬൫ோೄ_ು ାோಸಿವ_ಶೂ ൯

Where:

ܸ஼௉ ൌ 10% ‫்ܸ ݂݋‬ாௌ்_௉ ൌ 80ܸ
‫ݐ‬௦௪ଶ ൌ 245 ݉‫ܿ݁ݏ‬
ܴௌ_௉ ൌ 10݇Ω

Therefore, ‫ܥ‬௉ ൎ 825ߤ‫ܨ‬

ܴீே஽_ாொ ൌ 120 ݇Ω

A similar set of calculations can be performed for selecting ‫ܥ‬ே .

Now, using an arbitrarily low value for ‫ݐ‬ௌௐଵ ൌ 99 ݉‫( ܿ݁ݏ‬leaving 1 msec as

deadtime between ‫ݐ‬ௌௐଵ and ‫ݐ‬ௌௐଶ ), and assuming a level of charge of 98% for ‫ܥ‬௉ , ܴௌଷ
may be calculated as follows:

Where:

ܸ஼௉ሺ௧ୀ௧మ ሻ ൌ ்ܸாௌ்_௉ ‫ڄ‬

௧
ି ௌௐଵ
஼ು ‫ڄ‬ோೄయ ሻ
ሺ1– ݁

ܸ஼௉ ൌ 98% ‫ ܸ008 ݂݋‬ൎ 780ܸ
‫ݐ‬ௌௐଵ ൌ 99 ݉‫ܿ݁ݏ‬

Therefore, ܴௌଷ ൎ 32.5݇Ω

‫ܥ‬௉ ൌ 825‫ܨݑ‬

67

Since ‫ݐ‬ௌௐଵ ൌ ‫ݐ‬ௌௐଷ ൌ 99 ݉‫ , ܿ݁ݏ‬and ‫ݐ‬ௌௐଶ ൌ ‫ݐ‬ௌௐସ ൌ 245 ݉‫ , ܿ݁ݏ‬the total

period for leak detection in this case is ~700 msec. This is shown in Figure 41.

Figure 41: Sequencing of switches for simulation

For simplification of the analysis process, only the response characteristic of ܸ஼௉

will be evaluated throughout this section. Since all the data required for calculation of the
common mode resistance (ܴீே஽_ாொ ) can be gleaned from phase 2 of the diagnostic

cycle (i.e. ‫ݐ‬ௌௐଶ ), emphasis will be placed on evaluating the circuit’s response during this

phase.

TEST CASE 1
As described in section 4.2.2, this test case is designed to evaluate the response of
the ground fault detection system in an ideal, no fault state. The conditions and

68

constraints for the test are the same as in section 4.2.2. The results are plotted in Figure
42.

Figure 42: Simulation results for test case 1 of method 3

As is evident in Figure 42, the bottom graph represents the voltage ܸ஼௉ , the

middle graph represents the state of SW2 (2V being ON, and 0V being OFF), and the top

graph represents the voltage ܸ௉ாொ (i.e. ‫ܸܪ‬௉ with respect to chassis ground). Since

isolation between the high voltage bus and chassis ground is high, the discharge of
capacitor ‫ܥ‬௉ is shallow, and the equivalent ground resistance can be calculated as
follows:

ܸ஼௉ ሺ‫ ݐ‬ൌ ‫ݐ‬஼௉ ሻ ൌ ൫்ܸாௌ்_௉ – ܸ௉_ாொ ൯ ൈ ݁

Therefore,

69

ି

௧ௌௐଶ
஼ು ൫ோೄ_ು ାோಸಿವ_ಶೂ ൯

൅ ܸ௉_ாொ

ܴீே஽_ாொ ൌ
ܴீே஽_ாொ ൌ

‫ݐ‬௦௪ଶ
െ ܴௌ_௉
்ܸாௌ்_௉ – ܸ௉_ாொ
݈݊ ൬
൰ ‫ܥ‬௉
ܸ஼௉ – ܸ௉_ாொ

0.245
െ ሺ10 ‫01 ڄ‬ଷ ሻ
798.21– 315.95
ቁ ൈ 0.825 ൈ 10ି଺
݈݊ ቀ
751.92– 315.95
ܴீே஽_ாொ ൎ 2.96‫ܯ‬Ω

Since the threshold to set a ground fault is ்ܴு = 120k , no fault is set in this case.
TEST CASE 2
The second test case is to evaluate the response of the circuit when a non-isolated
single phase inverter is connected to the HV battery. The conditions and constraints are
similar to those described in section 4.2.2, with the exception that the DC-DC Converter
and Auxiliary Load 1 are also connected (thereby adding 100nF each of common mode

capacitance between ‫ܸܪ‬௉ and ground, and ‫ܸܪ‬ே and ground). The added 120Hz ripple

from the single phase inverter has a negligible impact on the voltage waveform ܸ஼௉ , as
shown in Figure 43. The level of discharge is higher than it was in test case 1 due to the
added common mode capacitance, as well as the 2M

70

resistors in the inverter.

Figure 43: Simulation results for test case 2 of method 3

By substituting the measured values into the equation for ܴீே஽_ாொ , we obtain

460k . The actual resistance to ground is fairly higher than 460k

(i.e. ~1.75M ), but

due to the presence of Y-caps in the DC-DC converter and Auxiliary load there is
additional energy drawn from ‫ܥ‬௉ to charge the Y-caps that were originally at ܸ௉_ாொ =

267.66V. This condition is representative of test case 5 of section 4.2.2, and could
potentially reduce the reliability of this GFDS. Since the diagnostic algorithm does not
consider Y-capacitance, a drop in voltage ܸ஼௉ due to this capacitance is interpreted as
reduced isolation between HV power net and chassis ground.
TEST CASE 3
This test case is to evaluate the response of the circuit in a true fault condition.
The conditions and constraints of this test case are similar to those in Test Case 3 of
section 4.2.2. The results of the simulation are shown in Figure 44.
71

Figure 44: Simulation results for test case 3 of method 3

The depth of discharge of ‫ܥ‬௉ is significant in this particular case, as expected.

Also, ܸ௉_ாொ almost approaches 0V, which is indicative that there is a huge imbalance in
the common mode impedance and voltage on the HV power net. Using the measured
values to calculate ܴீே஽_ாொ we obtain ~ 112.5k , which is consistent with the true
leakage condition in the system.
TEST CASE 4
The conditions and constraints of this test case are similar to those of test case 6 in

section 4.2.2. The purpose of this simulation is to evaluate the response of the leak
detection system to an AC ground fault of 1
Figure 45 below.

72

(short circuit). The results are shown in

Figure 45: Simulation results for test case 4 of method 3
Due to the existence of a short circuit path between one of the AC output lines
and chassis ground, there is a significant amount of AC voltage that is super imposed
onto the common mode DC bus; as explained in section 4.2.2. This can be seen in the top
graph of Figure 45. The superimposition of AC voltage onto the common mode DC bus

results in measurement signals ܸ஼௉ and ܸ௉ாொ being noisy and hence reduce fidelity in
the measured data. The reliability of the diagnostic circuit is therefore significantly
minimized in AC fault conditions.
A good example of this can be seen in the simulation above, where substitution of

the measured values into the equation for ܴீே஽_ாொ will result in a non-real value, due to

ܸ௉_ாொ being higher than ܸ஼௉ .

73

TEST CASE 5
The conditions and constraints of this test case are similar to those of test case 7 in
section 4.2.2. The purpose of this simulation is to evaluate the impact of voltage swings

in the DC bus on the current waveforms ‫ܫ‬ଵ and ‫ܫ‬ଶ . The results are shown in Figure 46
below.

Figure 46: Simulation results for test case 5 of method 3
Variation in the differential bus voltage does not significantly impact the common
mode bus voltages, as evidenced by the graphs shown in Figure 46. A 30V swing in the
high voltage bus (from t = 5 to t = 5.6) causes a negligible change in ܸ஼௉ (< 2V).

In summary, the GFDS discussed in this section performs the intended function in
systems that have fixed components whose ideal common mode impedance is known.
Further, distinction between a discharge caused by resistive elements vs. capacitive

74

elements is not possible, hence skewing the data used for diagnostics and affecting the
reliability of fault detection.
Introducing a high voltage power supply on the common mode bus that has an

output voltage higher than ܸ஻஺்் may be detrimental to Y-caps within peripheral

devices connected to the HV bus (as described in section 3.3). Finally, the low frequency
high voltage waveform imposed on the common mode bus may also interfere with or
affect circuitry in peripheral devices that are connected to the HV bus. This waveform is
shown in Figure 47 below.

Figure 47: Common mode bus voltage as a result of method 3

75

4.5 Caterpillar Inc – Systems and Methods for
Electrical Leakage Detection
4.5.1 Technology Overview
Inventors Robert Wayne Lindsey and Jennifer Leah Lindsey from Caterpillar Inc,
Peoria, IL improved upon a popular impedance measurement technique referred to as the
Wheatstone bridge, which although is effective in determining low impedances to ground
in the HV power system, becomes unreliable in the case of a balanced leak fault [20]. The
relevance of this invention may be understood by briefly examining the principle behind
the Wheatstone bridge, and its limitations. Masaki Yogou from Sanyo Electric Co., Japan
illustrates the implementation of the Wheatstone bridge principle to detect electrical
leakage in a power supply as shown in Figure 48 [21].

Figure 48: schematic overview of typical Wheatstone bridge method [21]

76

In the figure, a High Voltage battery is connected to an inverter across the

positive and negative terminals. A separate GFDC unit comprising ܴ஼ଵ , ܴ஼ଶ , ܴଵ , ܴଶ ,

ܴଷ , ܴସ , ܴହ , two op-amps, and an LED is also connected to the HV bus. ܴ஼ଵ and ܴ஼ଶ

are voltage dividing resistors of equal magnitude that are connected to a common

reference ܸ஼ . ܴଵ and ܴସ are protection resistors of high value (e.g. 1M ), which are

connected in series to ܴଶ and ܴଷ respectively. ܴଶ and ܴଷ are detection resistors of lower

value (22k ) that are connected to chassis ground through an isolation resistor ܴହ (e.g.
6M ). Resistors ܴସ and ܴଷ , and ܴଵ and ܴଶ have common references ܸ஻ and ܸ
஺

respectively. By measuring Voltages ܸଵூே (i.e. ܸ െ ܸ஼ ) and ܸଶூே (i.e. ܸ஻ െ ܸ஼ ), and
஺

comparing the measurements to a threshold voltage ்ܸு , a ground fault may be
diagnosed.

Under ideal conditions, the isolation between ‫ܸܪ‬௉ or ‫ܸܪ‬ே and chassis ground

(i.e. ܴ଺ ) is significantly high (multiples of M

- e.g. 10M ). In this case, the currents ݅

and ݅ீ (as shown in Figure 49) may be calculated as follows:

Where

ሺܴଷ ൅ ܴସ ሻሺܴହ ൅ ܴ଺ ሻ
ܴଷ ൅ ܴସ ൅ ܴହ ൅ ܴ଺
ܸ௑ ൌ ܸ஻஺்் ൮
൲
ሺܴଷ ൅ ܴସ ሻሺܴହ ൅ ܴ଺ ሻ
൬ܴଵ ൅ ܴଶ ൅
൰
ܴଷ ൅ ܴସ ൅ ܴହ ൅ ܴ଺
ܴଵ ൌ ܴସ ൌ 1‫ܯ‬Ω

ܴଶ ൌ ܴଷ ൌ 22݇Ω
ܴହ ൌ 6‫ܯ‬Ω
77

ܴ଺ ൌ 10‫ܯ‬Ω

Therefore, ܸ௑ ൌ 290.72 ܸ

Vେ ൌ V୆୅୘୘ ൬

R େଶ
56kΩ
൰ ൌ 600 ൬
൰ ൌ 300V
56kΩ ൅ 56kΩ
R େଵ ൅ R େଶ

Figure 49: Partial equivalent circuit of Wheatstone bridge [21]
Now,

݅ ൌ

݅ீ ൌ
Hence,

ܸ௑
ൌ 290.14ߤ‫ܣ‬
ܴଷ ൅ ܴସ
ܸ௑
ൌ 18.17ߤ‫ܣ‬
ܴହ ൅ ܴ଺

ܸଵ ൌ ሺ݅ ൅ ݅ீ ሻ ‫ܴ ڄ‬ଶ ൌ 6.78ܸ
ܸଶ ൌ ݅ ‫ܴ ڄ‬ଷ ൌ 6.38ܸ

ܸଵூே ൌ ܸ஺ െ ܸ஼ ൌ ሺܸ௑ ൅ ܸଵ ሻ െ ܸ஼ ൌ 290.72 ൅ 6.78 െ 300 ൌ െ2.5ܸ

ܸଶூே ൌ ܸ஻ െ ܸ஼ ൌ ሺܸ௑ െ ܸଶ ሻ െ ܸ஼ ൌ 290.72 െ 6.38 െ 300 ൌ െ15.66ܸ
78

Similarly, In the case of a true leak condition where ܴ଺ ൌ 100݇Ω, ܸଵூே is

calculated to be -16.24V and ܸଶூே to be -29.16V. In the case where the fault occurs
between ‫ܸܪ‬௉ and chassis ground, the same result as described above is obtained. Hence,
the threshold for a ground fault may be selected as ܸଵூே ൏ െ16ܸ or ܸଶூே ൐ 16ܸ .

One challenge with using such a method for ground fault detection is that a
compromise of isolation on BOTH the positive conductor to chassis and negative
conductor to chassis can skew the values of ܸଵூே and ܸଶூே such that they will NOT
trigger a fault. This is shown in section 4.5.2.

As mentioned earlier, the method developed by Robert and Jennifer Lindsey
operates on similar principles, but is modified to allow detection of balanced leak faults.

Figure 50: Schematic overview of method 4 [20]

79

Figure 50 shows the architecture of the GFDC [20]. In the figure, ܴଵ and ܴଶ are

two resistors of equal resistance providing means to balance and reference the high
voltage buses ‫ܸܪ‬௉ and ‫ܸܪ‬ே equally to chassis (or frame) ground (synonymous to ܴ஼ଵ

and ܴ஼ଶ described earlier in the section). ܴଷ and ܴସ form a voltage divider that is used
to measure ܸ௉ைௌ and ܸோீ in reference to chassis ground. The inherent leakage

resistance between HV+ and chassis, and HV– and chassis is shown as ܴ஻௅ଵ and ܴ஻௅ଶ

respectively. When ܴ஻௅ଵ and ܴ஻௅ଶ are equal in magnitude, the system is considered to

be in a balanced condition. Conversely, when ܴ஻௅ଵ does not equal ܴ஻௅ଶ , the system is

considered to be in an unbalanced condition, and the voltage across ‫ܸܪ‬௉ and ground

does not equal the voltage across ‫ܸܪ‬ே and ground. Therefore, the Imbalance of the
system may be defined by the following equations:

ܸ஻஺்் ൌ ‫ܸܪ‬௉ – ‫ܸܪ‬ே

‫ ݈ܾ݁ܿ݊ܽܽ݉ܫ‬ൌ 1 – ቌ

‫ܸܪ‬ே
ቍ
ܸ஻஺்்
2

Since ܸ௉ைௌ and ܸோீ refer to the voltage measurements across resistors ܴସ , and

can be represented in terms of ‫ܸܪ‬௉ and ‫ܸܪ‬ே , the expression for voltage imbalance may

be rewritten as:

‫ܸܪ‬௉ ൌ ܸ௉ைௌ ‫ڄ‬

80

ܴଷ ൅ ܴସ
ܴସ

‫ܸܪ‬ே ൌ ܸோீ ‫ڄ‬
Therefore,

ܴଷ ൅ ܴସ
ܴସ

ܸோீ
ܸ
െ ܸோீ
‫ ݈ܾ݁ܿ݊ܽܽ݉ܫ‬ൌ 1 െ ተ ௉ைௌ
ተ
2

For unbalanced fault conditions, the measured values of ܸ௉ைௌ and ܸோீ may be

used to calculate the degree of imbalance, and subsequently the magnitude of leakage
resistance. Assuming a ground fault between high voltage positive and chassis ground
(i.e. ܴ஻௅ଵ is a low value), ܸ௉ைௌ and ܸோீ maybe used to calculate ܴ஻௅ଵ .

ܸ௉ைௌ ൌ ሺ ܸ௉ைௌ െ ܸோீ ሻ ‫ ڄ‬ቆ

Where

ܴ௉ ൌ

ܴாொ_௉
ቇ
ܴாொ_௉ ൅ ܴாொ_ே

ܴଵ ‫ ڄ‬ሺܴଷ ൅ ܴସ ሻ
ܴଵ ൅ ܴଷ ൅ ܴସ

ܴாொ_௉ ൌ

ܴ஻௅ଵ ‫ܴ ڄ‬௉
ܴ஻௅ଵ ൅ ܴ௉

Now, assuming ܴ஻௅ଶ is very high (e.g. 10M )

ܴாொ_ே ൌ

Therefore,

ܴ஻௅ଶ ‫ܴ ڄ‬௉
ൎ ܴ௉
ܴ஻௅ଶ ൅ ܴ௉

ܴ஻௅ଵ ൌ െܴ௉ ‫ ڄ‬൬

ܸ௉ைௌ
൰
ܸ௉ைௌ ൅ ܸோீ

81

However, in a balanced fault condition, ܸ௉ைௌ and ܸோீ may show that the circuit

is balanced and the true leakage resistance may not be accurately calculated. In order to
detect the leakage resistance in a balanced fault condition, leakage detection resistor ܴ஽

may be switched into the circuit, as shown in Figure 50. For this purpose, any transistor
switch may be used, controlled by a signal issued by the microcontroller. When the

switch is ON, resistor ܴ஽ provides an additional current path from ‫ܸܪ‬௉ to chassis
ground, thus changing the equivalent resistance ܴாொ_௉ between high voltage positive rail

and chassis, and subsequently generate an offset voltage to unbalance the system.

ܴாொ_௉௡௘௪ ൌ

ܴாொ_௉ ‫ܴ ڄ‬஽
ܴாொ_௉ ൅ ܴ஽

Furthermore, the resistance of the parallel configuration of ܴଶ , (ܴଷ ൅ ܴସ ), and

ܴ஻௅ଶ (i.e. ܴாொ_ே௡௘௪ ) also equals that of ܴாொ_௉௡௘௪ because ܴ஻௅ଵ equals to ܴ஻௅ଶ , and

ܴଵ equals ܴଶ . Therefore, when ܴ஽ is switched into the system, the voltage distribution
across ‫ܸܪ‬௉ and ground, and ‫ܸܪ‬ே and ground can be expressed as follows:

ܴாொ_௉௡௘௪ ൌ ܴாொ_ே௡௘௪ ൌ ܴாொ_௡௘௪
ܴாொ_௉ ൌ ܴாொ_ே ൌ ܴாொ

Hence,

ܸ௉ைௌ ܴாொ_௡௘௪
ൌ
ܸோீ
ܴாொ

ܴாொ ൌ ܴ஽ ‫ ڄ‬ฬ

ܸோீ െ ܸ௉ைௌ
ฬ
ܸ௉ைௌ

82

In this manner, by switching ܴ஽ into the system, ܴாொ may be calculated followed by

ܴ஻௅ଵ and ܴ஻௅ଶ .

4.5.2 Simulation and Analysis
Although the mechanism used to calculate the leakage resistance in a balanced fault
condition is a unique modification to the original method, it uses the same Wheatstone
bridge principle. Hence, for simplicity of analysis, only the original method discussed for
detecting ground faults in an unbalanced condition will be simulated to verify reliable
operation under various test cases. This is also done in order to be consistent with the test
conditions used for analysis of other design methodologies. Figure 51 below shows the
circuit model used for simulation.

Figure 51: Leak detection circuit model for method 4

83

As mentioned, the resistor ܴ஽ and the transistor switch have not been modeled. In

addition, although the invention disclosure uses a value of 8k
resistors ܴଵ and ܴଶ , 10M

for the voltage divider

has been used for the simulation to in order to maintain large

isolation between the high voltage rails and chassis. The detection resistors ܴସ_௉ and

ܴସ_ே are 10k each in order to reduce the measured common mode voltage by a factor

of 200 so that it may be used as a voltage input for a microcontroller channel that is
limited to 5V.
TEST CASE 1
This test case is designed to evaluate the response of the ground fault detection
system in an ideal, no fault state. The conditions and constraints for the test are the same
as in section 4.2.2. The results are plotted in Figure 52.

Figure 52: Simulation results for test case 1 of method 4

84

As expected for a no-fault balanced system, the following measurements were realized:

ܸ௉ைௌ ൌ ܸ஻஺்் ‫ ڄ‬ቆ

ܴாொ_௉
ܴସ
ቇ
‫ڄ‬
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ

ܸோீ ൌ െܸ஻஺்் ‫ ڄ‬ቆ
ܸ௉ைௌ ൌ െܸோீ

ܴாொ_ே
ܴସ
‫ڄ‬
ቇ
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ

ܴாொ_௉ ൌ ܴாொ_ே

1
0.01 ‫଺01 ڄ‬
ൌ 600 ‫ ڄ‬൬ ൰ ‫ ڄ‬ቆ
ቇ ൌ 1.5ܸ
2
1.99 ‫଺01 ڄ‬

ܸோீ
ܸ
െ ܸோீ
% ‫ ݈ܾ݁ܿ݊ܽܽ݉ܫ‬ൌ 1 െ ተ ௉ைௌ
ተ ‫001 ڄ‬
2

െ1.5
1.5 െ ሺെ1.5ሻ
% ‫ ݈ܾ݁ܿ݊ܽܽ݉ܫ‬ൌ 1 െ ተ
ተ ‫ 001 ڄ‬ൎ 0%
2

In this case, since the imbalance in the circuit is zero, ܴ஽ could be switched into

the circuit to test for the leakage resistance in the common mode bus and verify that the
isolation is indeed high.
TEST CASE 2

The second test case is to evaluate the response of the circuit when a non-isolated
single phase inverter is connected to the HV battery. The conditions and constraints are
similar to those described in section 4.2.2. The added 120Hz ripple from the single phase

85

inverter has a negligible impact on the voltage waveforms ܸ௉ைௌ and ܸோீ , as shown in
Figure 53.

Figure 53: Simulation results for test case 2 of method 4
120Hz ripple is noticeable in the calculated waveform for system imbalance, but does not
have any consequential impact on fault diagnostics. A positive imbalance of ~18% shows

us that there is lower isolation resistance between ‫ܸܪ‬ே and chassis. Since this imbalance
is inherent to the setup of the circuit model for the single phase inverter, it is expected.
TEST CASE 3
This test case is to evaluate the response of the circuit in a true fault condition.
The conditions and constraints of this test case are similar to those in Test Case 3 of
section 4.2.2. The results of the simulation are shown in Figure 54.

86

Figure 54: Simulation results for test case 3 of method 4
When the 110k

resistor is induced between high voltage positive and chassis at t

= 3.0 s, both ܸ௉ைௌ and ܸோீ drop by 1.275 V. Hence, the imbalance in the network is
85%, and since 120k

is the threshold leakage resistance ்ܴு for the circuit, 80% (with

a 5% margin) may be used as the imbalance threshold for detecting a ground fault (i.e.
Imbalance > 80% will trigger a leak fault). However, this threshold is relevant only for a
system with the conditions similar to those that were used for this simulation, i.e. only the
traction inverter is connected to the HV bus and nothing else. If other known components
were introduced to the system that would imbalance the system further, the threshold
would need to change. This is discussed further in test case 7.
TEST CASE 4
This test case is synonymous to Test Case 5 in section 4.2.2, where the response
of the leak detection circuit to added Y-capacitance on the common mode bus is
evaluated. The results are shown in Figure 55.
87

Figure 55: Simulation results for test case 4 of method 4

The addition of 50nF of capacitance to the common mode bus at time ‫ ݐ‬ൌ

2.0 ‫ ܿ݁ݏ‬when switches S7 and S8 are closed does not have a significant impact on the

voltages ܸ௉ைௌ and ܸோீ since the overall common mode impedance only changes

momentarily as the capacitors charge up. A similar response is observed again at

‫ ݐ‬ൌ 3.0 ‫ ܿ݁ݏ‬when an additional 50nF of capacitance is added to the common mode bus,

before the system eventually reaches steady state at ‫ ݐ‬ൎ 4 ‫ .ܿ݁ݏ‬The level of voltage

imbalance and time taken for the voltage to stabilize is dependant on the magnitude of Ycapacitance in the system.
TEST CASE 5

The conditions and constraints of this test case are similar to those of test case 6 in
section 4.2.2. The purpose of this simulation is to evaluate the response of the leak

88

detection system to an AC ground fault of 1

(short circuit). The results are shown in

Figure 56 below.

Figure 56: Simulation results for test case 5 of method 4
As described in section 4.2.2, an AC leak fault within the HEV system causes a
large amplitude of ripple voltage to be superimposed on the DC side common mode bus,
typically at a frequency that is two times the fundamental frequency of the output of a
single phase inverter, and six times that of a three phase inverter (e.g. traction inverter).

This causes the voltages ܸ௉ைௌ and ܸோீ to oscillate at the same frequency, as illustrated

in the bottom graph of Figure 56. Consequently, the imbalance of the system also varies
at the same frequency. Depending on the sampling frequency of the voltage monitoring

device in the GFDS, certain algorithms may be implemented to identify an AC leak fault
in the system.

89

TEST CASE 6
The conditions and constraints of this test case are similar to those of test case 7 in
section 4.2.2. The purpose of this simulation is to evaluate the impact of voltage swings
in the DC bus on the voltage waveforms ܸ௉ைௌ and ܸோீ . The results are shown in

Figure 57 below.

Figure 57: Simulation results for test case 6 of method 4

Although variations in the differential voltage of the HV bus will impact the

voltages ܸ௉ைௌ and ܸோீ , the percentage of voltage imbalance will not change because
both positive and negative common mode voltages will increase or decrease by the same
factor. This may be explained mathematically as follows:
Recall that ܸ௉ைௌ and ܸோீ are defined as:
90

ܸ௉ைௌ ൌ ܸ஻஺்் ‫ ڄ‬ቆ

ܴாொ_௉
ܴସ
‫ڄ‬
ቇ
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ

ܸோீ ൌ െܸ஻஺்் ‫ ڄ‬ቆ

ܴாொ_ே
ܴସ
‫ڄ‬
ቇ
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ

Then, the change in these voltages i.e. ∆ܸ௉ைௌ and ∆ܸோீ is given by:

∆ܸ௉ைௌ ൌ ∆ܸ஻஺்் ‫ ڄ‬ቆ

ܴாொ_௉
ܴସ
‫ڄ‬
ቇ
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ

∆ܸோீ ൌ െ∆ܸ஻஺்் ‫ ڄ‬ቆ

ܴாொ_ே
ܴସ
‫ڄ‬
ቇ
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ

Therefore, assuming that the resistances of the circuit do not change,

∆ܸ௉ைௌ ܸ௉ைௌ
ൌ
∆ܸோீ ܸோீ
For a balanced circuit,

Hence, ∆ܸ௉ைௌ = ∆ܸோீ .

ܴாொ_௉
ܴସ
‫ڄ‬
൰
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ
ൌെ
ܴாொ_ே
ܴସ
‫ڄ‬
൬
൰
ܴாொ_௉ ൅ ܴாொ_ே ܴସ ൅ ܴଷ
൬

∆ܸ௉ைௌ ܸ௉ைௌ
ൌ
ൌ െ1
∆ܸோீ ܸோீ

TEST CASE 7
A unique test case was developed for this particular GFDS to evaluate its
effectiveness in detecting a fault in a “partially unbalanced” system. Two distinct fault
conditions (balanced and unbalanced) were discussed and methods to detect leakage

91

resistance under these conditions were proposed. However, a required condition for

effectively calculating ܴ஻௅ଵ or ܴ஻௅ଶ in an unbalanced fault state was that at least one of

these parameters is known (i.e. no isolation breakdown between positive or negative rail
and chassis).
Revisiting test case 3, if a 120k

fault was introduced between ‫ܸܪ‬௉ and chassis

ground at t = 3 sec, an imbalance of 85% would occur, and a leak fault would be

triggered. If a similar fault was introduced between ‫ܸܪ‬ே and chassis ground at t = 5 sec,
a balanced fault condition would exist, and ܴ஽ could be switched into the circuit to
determine ܴாொ and eventually ܴ஻௅ଵ (=ܴ஻௅ଶ ).

Instead, if the equivalent resistance between ‫ܸܪ‬ே and chassis was partially

reduced to ܴாொே ൌ 1‫ܯ‬Ω at t = 5 sec, an unbalanced fault condition would continue to

exist but the percentage imbalance would be different. Such an occurrence could be
caused by, for example, connecting an additional component to the flexible HEV bus that
has reduced isolation to ground. A simulation of this case is shown in Figure 58 on Page
93.

92

Figure 58: Simulation results for test case 7 of method 4

By reducing ܴாொ_ே from 2M

to 1M , the percentage imbalance in the circuit

decreased from 85% to 68%, in spite of the leakage resistance between ‫ܸܪ‬௉ and chassis
being only 120k . In this case, a leak fault would not be set, rendering both methods
only partially effective in detecting ground faults.
In summary, the GFDS discussed in this section is pretty robust in detecting
unbalanced and balanced leak faults; where an unbalanced leak fault is defined as
isolation breakdown between either positive rail and chassis, or negative rail and chassis,
but not both. However, this GFDS is not capable of accurately detecting low isolation in
the case of a partially unbalanced fault, as described in test case 7. Also, it has been
shown that ripple voltage or noise on the common mode bus can skew measurements of

ܸ௉ைௌ and ܸோீ . Finally, although it is important to keep the value of ܴ஽ low (e.g.
93

10k ) in order to accurately detect low isolation resistance in a balanced fault condition,
this component creates an additional leakage path to chassis by significantly reducing the
common mode equivalent impedance in the system during the transistor switch ON
period.
One recommendation for improving the GFDS would be to add switches SW1

and SW2 between ܴଷ_௉ and ܴସ_௉ , and ܴଷ_ே and ܴସ_ே respectively. Similar to the

GFDS analyzed in section 4.3, switches SW1 and SW2 would be controlled
independently and switched at alternate times so that both switches are never ON at the
same time. One embodiment of this is shown in Figure 59 below.

Figure 59: Circuit improvement for method 4

ܴாொ_௉ and ܴாொ_ே are the equivalent resistances between high voltage positive or

negative with respect to ground, and ‫ܸܪ‬௉ and ‫ܸܪ‬ே represent high voltage positive and

negative bus voltages respectively. Also, balancing resistors ܴଵ and ܴଶ may be
94

eliminated since they do not contribute to the calculation of ܴாொ_௉ and ܴாொ_ே as

described below.
Assuming

ܴଷ_௉ ൌ ܴଷ_ே ൌ ܴଷ ൌ 1.99‫ܯ‬Ω,

0.01‫ܯ‬Ω, when SW1 is first turned ON,

ܸଵ ൌ ‫ܸܪ‬௉ ‫ ڄ‬൬

Therefore,

ܸଵ ൌ

and

ܴସ_௉ ൌ ܴସ_ே ൌ ܴସ ൌ

ܴସ
൰
ܴଷ ൅ ܴସ

ܴாொ_௉ ‫ ڄ‬ሺܴଷ ൅ ܴସ ሻ
ܴாொ_௉ ൅ ܴଷ ൅ ܴସ
‫ܸܪ‬௉ ൌ ܸ஻஺்் ‫ڄ‬
ܴாொ_௉ ‫ ڄ‬ሺܴଷ ൅ ܴସ ሻ
ܴாொ_ே ൅ ൬
൰
ܴாொ_௉ ൅ ܴଷ ൅ ܴସ
ܸ஻஺்் ‫ܴ ڄ‬ସ ‫ܴ ڄ‬ாொ_௉
ܴாொ_௉ ‫ܴ ڄ‬ாொ_ே ൅ ሺܴଷ ൅ ܴସ ሻ ‫ܴ ڄ‬ாொ_௉ ൅ ሺܴଷ ൅ ܴସ ሻ ‫ܴ ڄ‬ாொ_ே

Similarly, when SW1 is OFF and SW2 is ON,

ܸଶ ൌ ‫ܸܪ‬ே ‫ ڄ‬൬

ܸଶ ൌ

ܴସ
൰
ܴଷ ൅ ܴସ

ܴாொ_ே ‫ ڄ‬ሺܴଷ ൅ ܴସ ሻ
ܴாொ_ே ൅ ܴଷ ൅ ܴସ
‫ܸܪ‬ே ൌ ܸ஻஺்் ‫ڄ‬
ܴாொ_ே ‫ ڄ‬ሺܴଷ ൅ ܴସ ሻ
ܴாொ_௉ ൅ ൬
൰
ܴாொ_ே ൅ ܴଷ ൅ ܴସ

ܸ஻஺்் ‫ܴ ڄ‬ସ ‫ܴ ڄ‬ாொ_ே
ܴாொ_௉ ‫ܴ ڄ‬ாொ_ே ൅ ሺܴଷ ൅ ܴସ ሻ ‫ܴ ڄ‬ாொ_௉ ൅ ሺܴଷ ൅ ܴସ ሻ ‫ܴ ڄ‬ாொ_ே

Now, solving the expressions for ܸଵ and ܸଶ for ܴாொ_௉ , we get
95

ܴாொ_௉ ൌ െ
And,

ሺܴଷ ൅ ܴସ ሻ ‫ܴ ڄ‬ாொ_ே ‫ܸ ڄ‬ଵ
ܴଷ ‫ܸ ڄ‬ଵ ൅ ܴସ ‫ ڄ‬ሺܸଵ െ ܸ஻஺்் ሻ ൅ ܴாொ_ே ‫ܸ ڄ‬ଵ

ܴாொ_௉ ൌ െ

൫ܴଷ ‫ܸ ڄ‬ଶ ൅ ܴସ ‫ ڄ‬ሺܸଶ െ ܸ஻஺்் ሻ൯ ‫ܴ ڄ‬ாொ_ே
൫ܴଷ ൅ ܴସ ൅ ܴாொ_ே ൯ ‫ܸ ڄ‬ଶ

Finally, equating these two expressions for ܴாொ_௉ , and solving for ܴாொ_ே , we get

ܴாொ_ே ൌ െ

൫ܴଷ ‫ ڄ‬ሺܸଵ ൅ ܸଶ ሻ ൅ ܴସ ‫ ڄ‬ሺܸଵ ൅ ܸଶ െ ܸ஻஺்் ሻ൯
ܸଵ

ܴாொ_௉ ൌ െ

൫ܴଷ ‫ ڄ‬ሺܸଵ ൅ ܸଶ ሻ ൅ ܴସ ‫ ڄ‬ሺܸଵ ൅ ܸଶ െ ܸ஻஺்் ሻ൯
ܸଶ

This process may be repeated for ܴாொ_௉ to arrive at

The performance of this modified GFDS would be very similar to that of section
4.3 due to similarity in the operation principle.

96

CHAPTER 5
5 Comparison of Technologies
Although a small selection of existing technologies have been reviewed, it is
important to understand that there are many other variations of ground fault detection
systems designed for a plethora of industries and applications. Ultimately, all ground
fault detection systems are developed to identify the impedance between a high voltage
conductor and ground, and this can fundamentally be achieved in a finite number of
ways.
The four methods that were evaluated in chapter 4 had some fundamental
differences in method of operation and thus were chosen to be reviewed. In order to
benchmark these methodologies against an ideal GFDS, a popular engineering method
known as Pugh analysis will be used that helps determine which potential solutions are
better than others, based on pre-defined criteria. It is a scoring matrix used for concept
selection, in which options are assigned scores relative to criteria. The best approach is
selected based on the consolidated scores [22].

5.1 Criteria for Analysis and Weighting Factor
In order to proceed with the analysis, it is critical to identify important
characteristics of a GFDS that is to be implemented in a Flexible Hybrid Power System.
These characteristics may be used as criteria for comparison.

97

5.1.1 Detection time
The response time of a diagnostic circuit and its embedded software to a true
ground fault condition is a critical parameter to be considered when designing the GFDS.
UL specification 2231-2 states that a ground fault circuit interrupting device shall open a
ଵ.ସଷ

circuit in no more than ܶ ൌ ቆ20ൗ‫ ܫ‬ቇ

, where T is time in seconds, and I is the fault

trip current in mA (e.g. 5mA) [10]. Keeping in mind that this specification written for
electric charging systems in vehicles is very conservative, and that T = 7.25 sec is a
requirement for AC faults (for which limits are more stringent), it is perhaps acceptable
for trip times to be higher. Moreover, most standards such as UL 2231 are developed for
Ground Fault Interrupters which are meant to be reactive, instead of Vehicle Ground fault
Detection Systems which are meant to be proactive. As explained in section 3.2.2, loss of
isolation between a high voltage conductor and chassis ground only creates a shock
hazard, but does not cause a shock. We may still use T = 7.25 sec for defining various
ratings that a GFDS can be classified under.
Table 2: Definition of ratings for criteria 1
Rating

Description
System cannot reliably diagnose a leak fault in less than 7.25 sec, and
cannot be designed to perform continuous monitoring of the power system.

0
This may be because of limitations inherent to the design, or incompatibility
with the HEV power system
system can be designed to have a reliable detection time of less than 7.25
1
sec, but cannot perform continuous monitoring of the power system

98

system can perform continuous monitoring of the high voltage system for
2

leak faults, but cannot be designed to establish a true fault within 7.25 sec of
leakage occurrence
system can perform continuous monitoring of isolation faults and diagnose a

3
leak fault within 7.25 sec of occurrence

The weighting factor for this criterion may arbitrarily be selected as 4 (from a scale of 1
to 5), since fast detection time is critical to rapid mitigation of the negative effects of leak
faults as described in chapter 3.

5.1.2 Type of Fault
Although the GFDS is connected to the HVDC bus, it is expected to diagnose
leak faults that may exist anywhere in the power system, including those between AC
lines (that are not isolated from DC) and chassis ground. The weighting factor for this
criterion may arbitrarily be selected as 2 (from a scale of 1 to 5), since additional GFDSs
for AC leak faults may be implemented in the system if the GFDS being rated is not
capable of detecting both AC and DC faults.
Table 3: Definition of ratings for criteria 2
Rating
1
2

Description
System can detect DC faults only
System can detect both DC and AC faults

99

5.1.3 Mode of Detection
A flexible HEV power system can have a very complex architecture depending on
the nature of components connected to the HV bus. Hence, HV bus characteristics such
as current, voltage, and impedance may be very dynamic, creating the need for a GFDS
that can function reliably in all states of operation. These states include:
•

HEV Off mode – The hybrid power system is powered down

•

HEV Standby/ Stationary mode – The High Voltage system is live, but the vehicle
is not in motion, and all power electronic loads are disconnected. This can also be
considered electrical steady state.
HEV ON/ drive mode – The HV system is live, and devices such as the traction

•

inverter, DC-DC Converter, single phase inverter etc. are in conditioning power.
Hence, the ratings for this criterion may be defined as shown in Table 4. The
weighting factor for this criterion may arbitrarily be selected as 4 (from a scale of 1 to 5),
since it is critical that the system functions in all modes of operation of the HEV.
Table 4: Definition of ratings for criteria 3
Rating
0

Description
System can only perform diagnostics during the HEV OFF state
System can perform diagnostics when the high voltage system is live, but

1

only in an electrically steady state when the HEV is stationary and no power
electronic loads are being powered.

2

System can perform reliable diagnostics in any state of operation

100

5.1.4 Flexibility of Design
One of the most critical factors in developing a GFDS for a HEV with flexible
architecture (as described in section 4.1) is the ability of the system to adapt to changes in
the high voltage bus configuration, either intrinsically or through non-intrusive system
calibration that can be performed without impacting the design life cycle of the HEV
system or subsystem. Typically, if a GFDS requires a hardware design change in order to
function effectively in a new high voltage bus configuration/ application, it would not be
suitable for use in a flexible high voltage system. However, if all that was required was a
software calibration or a patch downloadable to the hybrid system controller through a
vehicle diagnostic tool, then adaptation becomes seamless. This will be discussed in
detail in section 5.2.
The ratings for this criterion may therefore be defined as shown in Table 5. The
weighting factor would be 5 (from a scale of 1 to 5) because of the significance of this
feature in a flexible power system.
Table 5: Definition of ratings for criteria 4
Rating

Description
System requires a hardware change to be compatible with a high voltage bus

0
configuration that varies from the original configuration it was designed for
System contains software that can be calibrated or modified to adapt to a
2
new high voltage bus configuration

101

5.1.5 Susceptibility to common noise sources
System monitoring devices and controllers in a HEV system are constantly
exposed to electromagnetic interference (EMI). Common mode conducted noise on the
high voltage bus and ripple current produced by the mutual interaction of power
converters with the battery can be of significant interference for voltage and current
monitoring circuits [23]. The fundamental frequency of this common mode voltage may
be as low as 60 Hz, but it can reach 120 kHz with considerable amplitudes as shown in
Figure 60. Other sources of interference can be electronic devices or controllers resident
on the common mode bus that employ current switching techniques to perform their
function (such as secondary isolation monitoring circuits). The interference caused by
noise sources in the HEV system could result in the GFDS not setting a leak fault when it
actually exists (Type I error), or conversely setting a leak fault when it actually does not
exist (Type II error). Both of these are undesired effects, for reasons described in section
3.1.

Figure 60: High frequency Common mode noise in PWM inverter (100V/div) [23]
Therefore, it is important to select a GFDS methodology that is fairly immune to
conducted and radiated emissions in the hybrid system. At the same time, it is also

102

preferable that the GFDS does not become a source for conducted emissions that could
interfere with the operation of other monitoring circuits in the system. Hence, the ratings
for this criterion may be defined as shown in Table 6, and the weighting factor may
arbitrarily be selected as 4 (from a scale of 1 to 5), since immunity to noise relates
directly to the functionality of the GFDS.
Table 6: Definition of ratings for criteria 5
Rating

Description
System is susceptible to interference, resulting in a Type I error or Type II

0
error or both
System is fairly immune to common mode noise, or is capable of
1

identifying an input signal of low fidelity and reporting a fault. However,
system can be a source for conducted emissions
System is fairly immune to common mode noise, or is capable of

1.5

identifying a diagnostic input signal of low fidelity and reporting a fault.
System does not generate noise.

5.1.6 Cost & Complexity
One observation that can be made about the systems evaluated in chapter 4 is that
most of the systems require few components to functionally operate. A majority of the
components being used are discrete resistors, capacitors, and switches. Almost all the
systems include a processor or microcontroller that has intelligence to performance
diagnostics based on inputs from digital sensors in the circuit. Some systems utilize
additional power supplies in order to fundamentally perform leakage detection. With a

103

variety of architectures, an important criterion for consideration is the actual cost and
overall reliability of the system.
In general, the cost and reliability of a system improves as the complexity and
number of components within the system is reduced. Using GFDS 2 (the system
described in section 4.3) as a benchmark, the ratings for this criterion may be defined as
shown in Table 7. The weighting factor may be arbitrarily selected as 1 (from a scale of 1
to 5), since this criteria is not necessarily correlated to the functional robustness of the
GFDS. NOTE: The rating selection for this criterion is highly simplified, and is arbitrary
at best.
Table 7: Definition of ratings for criteria 6
Rating
0

Description
System design is more complex than GFDS 2

1

System design is similar in complexity to GFDS 2

2

System design is less complex than GFDS 2

5.2 System Benchmarking
The next step after defining criteria for analysis is to rate each system against this
set of criteria.
CRITERIA 1: DETECTION TIME
Method 1: Nissan Motor Design
As mentioned in section 4.2, this GFDS was designed to perform continuous
monitoring of the high voltage system for ground faults. Assuming a very low sampling

104

frequency of ݂ ൌ 10 ‫ ݖܪ‬for the amplitude of ܸ஽ா் , and conservatively using ܰ ൌ 70 as

the number of consecutive counts of ܸ஽ா் being below the leak threshold ்ܸு in order

to trigger a leak fault, ܶ ൌ ቀ1ൗ݂ ቁ ‫ ܰ ڄ‬ൌ 7 ‫ ܿ݁ݏ‬the minimum time required for the system

to reliably detect a leak fault. In this case, method 1 would receive a rating of 3.

Usually, 1 or 2 data points are not sufficient for an accurate diagnosis of the state
of the system, because of potential noise in the signal that could skew data measurements.
Although one exemplary implementation of the method was used to rate the system, it is
not implied that the system cannot be optimized to perform leakage detection in a shorter
period of time.
Method 2: Maxim Inc. Design
Similar to method 1, using a sampling frequency of ݂ ൌ 1 ‫ ݖܪ‬for ‫ܫ‬ଵ and ‫ܫ‬ଶ , and

ܰ ൌ 7 as the number of consecutive counts of leakage current ‫ܫ‬௅௄ being greater than the

threshold current ‫்ܫ‬ு in order to trigger a leak fault, ܶ ൌ 7 ‫ ܿ݁ݏ‬is the minimum time

required for the system to reliably detect a leak fault. In this case, method 2 would
receive a rating of 3.
Method 3: Lear Corp Design
Using the model developed in section 4.4, the period to calculate one data point of

ܴீே஽_ாொ is t=700msec. Assuming N=10 as the number of consecutive counts of leakage
current ܴீே஽_ாொ being less than the threshold leakage resistance ்ܴு in order to trigger

a leak fault, T= 7 sec is the minimum time required for the system to reliably detect a

105

leak fault. Also, this design can be implemented for continuous monitoring of the system
for faults. Hence, method 2 would receive a rating of 3.
Method 4: Caterpillar Inc Design
Although this method described in section 4.5 may be used for continuous

monitoring of the system, if resistor ܴ஽ (e.g. 10k ) was to be switched into the system

to calculate ܴ஻௅ in a balanced fault condition, leakage detection would be less frequent.
Increasing the value of ܴ஽ would also increase the time to detect a leak fault because of
increased settling time for common mode voltage due to the presence of Y-capacitors.
Hence, this system would receive a rating of 1.
CRITERIA 2: TYPE OF FAULT
Method 1: Nissan Motor Design
Based on the evaluation of this method, unless a major re-design of the GFDS
was performed, it would only be capable of detecting DC faults i.e. rating 1.
Method 2: Maxim Inc. Design
As mentioned in section 4.3, an increase in sampling frequency of the GFDS and
modification to the diagnostic algorithm, method 2 would be capable of effective fault
detection on both AC and DC buses, i.e. rating 2.
Method 3: Lear Corp Design
As explained in section 4.4, this method would only be capable of detecting DC
ground faults, i.e. rating 1.
106

Method 4: Caterpillar Inc Design
Similar to method 1, if the sampling frequency of ܸ௉ைௌ and ܸோீ was sufficiently

increased, the GFDS could potentially detect both AC and DC faults, i.e. rating 2.
CRITERIA 3: MODE OF DETECTION
Method 1: Nissan Motor Design

Since this GFDS is susceptible to ripple voltage in the common mode bus, as well
as Y-capacitance in the high voltage system, it may only perform reliable diagnostics
when the system is in standby/steady state. Hence, it would be given a rating of 1.
Method 2: Maxim Inc. Design
This GFDS is fairly robust in any state of operation, therefore it would receive a
rating of 2.
Method 3: Lear Corp Design
As explained in section 4.4.2, added Y-capacitance in the system could affect the
effectiveness of GFDS by causing it to set a leak fault when the leakage resistance is not
actually below 120k . Hence, the GFDS may only perform reliable diagnostics in
standby mode (when all components are disconnected from the battery), resulting in
rating of 1.
Method 4: Caterpillar Inc Design
This GFDS is fairly robust in any state of operation, therefore it would receive a
rating of 2.

107

CRITERIA 4: FLEXIBILITY OF DESIGN
Method 1: Nissan Motor Design
When implemented in a HEV with a Y-capacitance footprint that is different from
what the GFDS was originally designed for, a hardware change would be necessary to
maintain diagnostic effectiveness. As explained in section 4.2.2, due to the principle of

operation of this GFDS, reducing ்ܸு or adding software filters will not be of any
benefit. Hence, the rating for the system would be 0.
Method 2: Maxim Inc. Design
Since this GFDS is fairly immune to common mode ripple voltage, and can
perform diagnostics reliably irrespective of the Y-capacitance in the system, a simple
software change for ݂
ௌௐଵ and ݂
ௌௐଶ should be sufficient to allow implementation in a

HEV with a different architecture. Hence, this method would receive a rating of 2.
Method 3: Lear Corp Design

Similar to method 1, the reliability of this GFDS would vary depending on the
architecture of the HEV system, and there are no software changes that can be made to
allow this GFDS to adapt to a different HEV system than it was designed for. Hence, this
method would receive a rating of 0.

108

Method 4: Caterpillar Inc Design
Similar to method 2, This GFDS is fairly adaptable to any HEV system
architecture with little to no software changes. Hence, this method would receive a rating
of 2.
CRITERIA 5: SUSCEPTIBILITY TO COMMON NOISE SOURCES
Method 1: Nissan Motor Design
This GFDS is most definitely susceptible to noise, as discussed in section 4.2.2.
Hence, it would receive a rating of 0.
Method 2: Maxim Inc. Design
Even though this GFDS is fairly immune to noise on the common mode bus,
software algorithms may be implemented to allow the system to identify low fidelity
measurements (large high frequency noise) and trigger a fault. However, since the
method of operation is based on forcing a voltage imbalance in the system (when SW1
and SW2 are switched), it could be a source of interference for other monitoring devices
in the system. Hence, this system would receive a rating of 1.
Method 3: Lear Corp Design
Similar to method 2, even though the system is fairly immune to common mode

noise, it causes significant shifts of voltage in the common mode bus (since ்ܸாௌ்_௉ and

்ܸாௌ்_ே are > ܸ஻஺்் ), potentially interfering with the operation of other circuits in the
system. Hence, this system would receive a rating of 1.

109

Method 4: Caterpillar Inc Design
Once again, although the system is fairly immune to common mode noise, due to
the GFDS’s inability to detect a DC ground fault that is “partially unbalanced” as
described in 4.5.2 (resulting in a Type I error), this system would receive a rating of 1.
CRITERIA 6: COST & COMPLEXITY
Method 1: Nissan Motor Design
Since the only components needed for this GFDS are capacitors, resistors, diodes,
and a few op-amps, the design complexity is comparable to GFDS 2, i.e. rating would be
1.
Method 2: Maxim Inc. Design
The rating of this system would be 1 by default.
Method 3: Lear Corp Design
Because this GFDS requires the use of 2 boost converters, the complexity and
cost of the system increases dramatically. It would receive a rating of 0.
Method 4: Caterpillar Inc Design
This detection method is fairly simple to implement, and is comparable in
complexity to GFDS 2. Therefore, the rating would be 1.

110

5.3 Pugh Analysis
Using the methods for leak detection, criteria for analysis, weighting factors for
individual criterion, and assigned rating per criterion for each method, the Pugh Matrix in
Table 8 is developed.
Table 8: Pugh Matrix
Ground Fault detection Technology
Critical Criteria

Weight

Method 1

Method 2

Method 3

Method 4

Detection Time

4

3

3

3

1

Type of Fault

2

1

2

1

2

Mode of Detection

4

1

2

1

2

Flexibility of Design

5

0

1

0

1

Susceptibility to Noise

4

0

1

1

1

Complexity & Cost

1

1

1

0

1

Total

19

34

22

26

Based on the Pugh matrix, it is evident that method 2 described in section 4.3 is
the best solution for a GFDS in a flexible HEV architecture.

111

CHAPTER 6
6 Conclusion
As Hybrid Electric Vehicles penetrate the commercial vehicle industry at an
extremely rapid pace, it has become important for manufacturers to develop flexible
power system architectures that adapt to a multitude of vehicular applications. Equally
important is the need for the hardware of subsystems and components of the flexible
power systems to be designed for compatibility with alien devices and loads that may be
connected to the power system. Four Ground Fault Detection Systems developed for
electric vehicles have been analyzed for theory of operation, and tested for functional
effectiveness using circuit modeling and simulation techniques.
Some of the major challenges that are faced by engineers designing GFDSs for
flexible HEVs are varying magnitudes of Y-capacitance in power electronics
components, and the unknown nature of frequency, phase, and amplitude of the voltage
ripple generated by them. Not only were the theories of operation verified, but specific
test cases were designed to evaluate the performance of the following four methods of
isolation detection against these issues:
•

Method 1: Determining the leakage resistance between high voltage conductors
and chassis ground by injecting a square wave pulse into the common mode bus
through a capacitively coupled path, and measuring the level of attenuation.

•

Method 2: Alternately switching a known resistance into the positive and negative
common mode buses and measuring the leakage current through the resistance to

112

calculate the total equivalent leakage resistance between high voltage and chassis
ground.
•

Method 3: Charging a fixed capacitance to a known high voltage, and
subsequently discharging the energy into the common mode bus through a fixed
resistance in order to measure the time constant of the circuit and eventually the
equivalent leakage resistance.

•

Method 4: Measuring the percentage of voltage imbalance between the high
voltage positive bus and chassis, and high voltage negative bus and chassis, and
subsequently comparing the measurement to a threshold value to detect isolation
breakdown.
Finally, a Pugh analysis was done to determine the most suitable method amongst

by comparing them against six criteria: detection time, type of fault detection, mode of
detection, flexibility of design, susceptibility to noise, and design complexity. Method 2
was determined to be the most robust solution due to the flexibility in design and the
immunity to noise, and this process successfully demonstrated that existing fault
detection theories could be used for implementation in flexible high voltage power
systems.
A plethora of detection strategies and methodologies exist in the industry today
with consistent incremental benefits being gained for every new invention. This has been
confirmed through a literature search on GFDSs for HEVs. However, a significant
portion of these methodologies share the same fundamental operating principles as
revealed in this study. Although most of the techniques have been presented in this paper,

113

there are methods that have not been researched, such as differential monitoring of float
current using split core transformers or half effect sensors. This could be a potential area
for future research work.
It has also been demonstrated that AC fault detection using hardware for DC
GFDSs can be achieved by modifying the leakage detection algorithm. There are several
papers and invention disclosures that propose potential methods for this purpose. Finally,
future work could also analyze techniques for GFDS self diagnostics (to prevent type I
and type II errors), methods to increase detection time, and algorithms to identify the
source of the ground fault.

114

BIBLIOGRAPHY

115

BIBLIOGRAPHY

[1] Kevin R. Wheeler, Dogan A. Timucin, I. Xander Twombly, Kai F. Goebel, and Phil
F. Wysocki, "Aging Aircraft Wiring Fault Detection Survey," NASA Ames
Research Center, Moffett Field, CA, June 1, 2007.
[2] Weijun Yin and Daniel Schweickart, "Dielectric Breakdown of Polymeric Insulation
Films Under AC, DC and Pulsed Voltages," in Proc. 2009 IEEE Electrical
Insulation Conference, Montreal, QC, Canada, 31 May - 3 June 2009, pp. 292-296.
[3] Edward P. Rafter, "DC Ground Fault Detection Provided for Uninterruptible Power
Supplies," in The Battcon 2006 Proceedings, Orlando, FL, 2006, pp. 1-8.
[4] Stephen McCluer, "Battery Technology for Data Centers and Network Rooms:
Lead-Acid Battery Options," APC by Schneider Electric, White Paper 30 2010.
[5] V. Serrao, A. Lidozzi, L. Solero, and A. Di Napoli, "Common and Differential Mode
EMI Filters for Power Electronics," in Proc. 2008 International Symposium on
Power Electronics, Electrical Drives, Automation and Motion, Ischia, 11-13 June
2008, pp. 918-923.
[6] G. Anderson, "Galvanic Isolation in UPS," GUTOR Electronic Ltd, Wettingen,
Switzerland, 2004.
[7] M. Sudha and P. Anbalagan, "A Protection Scheme for Three-Phase Induction
Motor from Incipient Faults Using Embedded Controller," Asian Journal of
Scientific Research, vol. 2, no. 1, pp. 28-50, 2009.
[8] Yoshihiro Murai, Takehiko Kubota, and Yoshihiro Kawase, "Leakage Current
Reduction for a High-Frequency Carrier Inverter Feeding an Induction Motor,"
IEEE Transactions on Industry Applications, vol. 28, no. 4, pp. 858-863, July/
August 1992.
[9] J.G. Webster, Medical Instrumentation: Application and Design, 3rd ed. New York,
USA: John Wiley & Sons, 1998.
[10] Underwriters Laboratories Inc. (UL), UL 2231-2 Standard For Personnel Protection
Systems for Electric Vehicle (EV) Supply Circuits: Particular Requirements for
Protection Devices for Use in Charging Systems, 1st ed., May 1, 2002.

116

[11] Supratim Basu and Tore. M. Undeland, "Voltage and Current Ripple Considerations
for Improving Lifetime of Ultra-Capacitors Used for Energy Buffer Applications at
Converter Input," in Proc. 2010 Applied Power Electronics Conference and
Exposition (APEC), Palm Springs, CA, 18 March, 2010, pp. 244-247.
[12] Rudy Smaling, "Integrated Energy Management System for Hybrid Commercial
Vehicles," Eaton Corporation, Presented at Conference - Elektrik und Elektronik im
Nutzfahrzeug October 13-14, 2010.
[13] Tsuyoshi Morita and Shinsuke Nakazawa, "Ground Fault Detector for Vehicle," US
Patent Application 2007/0008666 A1, January 11, 2007.
[14] Thomas Friedli and Johann W. Kolar, "Comprehensive Comparison of Three-Phase
AC-AC Converter and Voltage DC-Link Back-to-Back Converter Systems," in
Proc. 2010 International Power Electronics Conference, Sapporo, 21-24 June, 2010,
pp. 2789-2798.
[15] Lixin Tang and Gui-Jia Su, "Boost Mode Test of a Current-Source-Inverter-Fed
Permanent Magnet Synchronous Drive for Automotive Applications," in Proc. 2010
IEEE 12th Workshop on Control and Modeling For Power Electronics, Boulder,
CO, 28-30 June 2010, pp. 1-8.
[16] Srdjan M. Lukic, Sanjaka G. Wirasingha, Fernando Rodriguez, Jian Cao, and Ali
Emadi, "Power Management of an Ultracapacitor/ Battery Hybrid Energy Storage
System in an HEV," in Proc. 2006 IEEE Vehicle Power and Propulsion Conference,
Windsor, Canada, 6-8 Sept, 2006, pp. 1-6.
[17] Hans-George Schweiger et al., "Comparison of Several Methods for Determining the
Internal Resistance of Lithium Ion Cells," Sensors, vol. 10, pp. 5605-5625, June
2010.
[18] Mark Plagens and Brian Fritz, "Fault Detection Method For Detecting Leakage
Paths Between Power Sources and Chassis," US Patent Application 2010/0207635
A1, August 19, 2010.
[19] Miguel Angel Acena and Jordi Escoda, "Ground-Fault Detection System for
Vehicles with a High-Voltage Power Net," US Patent Application 2009/0278547
A1, November 12, 2009.
[20] Robert Wayne Lindsey and Jennifer Leah Lindsey, "Systems and Methods for
Electrical Leakage Detection," US Patent 7459914 B2, December 2, 2008.

117

[21] Masaki Yugou, "Circuit For Detecting Leakage in Power Supply," US Patent
6700384 B2, March 2, 2004.
[22] (2009) Process Management & Analysis. [Online].
http://www.processma.com/resource/pugh_matrix.htm
[23] Zbigniew Noworolski and Eric Roman, "Ripple, Noise, Charge and Discharge:
Turning Unwanted Signals for Your Advantage," in The Battcon 2008 Proceedings,
Marco Island, FL, 2008, pp. 1-6.

118