Overvoltage and Surge Protection in

Variable Frequency Drives

A Dissertation Presented
by
Dawood Talebi Khanmiri

to
The Department of Electrical and Computer Engineering

in partial fulfillment of the requirements

for the degree of

Doctor of Philosophy
in the field of
Electrical Engineering

Northeastern University
Boston, Massachusetts

October 2020
 Acknowledgment

I would like to thank my advisor, Professor Brad Lehman, for his continuous support of my PhD

research, for his patience and guidance, without whom this thesis could not be completed.

Besides my professor, I would like to thank Mersen who partially funded this research and generously

provided the lab for high power experiments, especially Roy Ball, Jerry Mosesian, and late Craig

McKenzie, who helped me during my research.

I would, also, like to thank my thesis committee members, for their time and their insightful comments

in my proposal review. I would like to thank all of my labmates and fellow grad students, too, for all the

good memories that we have had together.

Last, but not least, I would like to thank my family who supported me in my studies, especially my

wife, Dr. Nasibeh Nasiri, without whom I would not have been able to complete my PhD.

i
 Abstract

Overvoltage and surge protection of electric devices is an indispensable part of the system protection.

Metal Oxide Varistors (MOVs) are the most common components used for surge protection. However,

MOVs are known to degrade over time when they experience severe discharges. This degradation may,

particularly, lessen protection capabilities for fast switching PWM applications, such as inverters and

drives. For instance, Variable Frequency Drives (VFDs) are becoming more and more widespread for

motor speed control and energy saving purposes. However, when connecting long cables between the

inverter and the motor, high frequency overvoltages appear on motor terminals due to the voltage reflection

phenomenon. These overvoltages can damage the motor and the cable insulation, and at the same time,

cause installed MOVs to conduct high currents and fail. In this research, we investigate the MOV’s

behavior and degradation in such applications. The VFD systems and characteristics of the overvoltages are

first modeled. MOVs behavior is investigated through experimental tests and simulations. Finally, new

methods of applying MOVs in PWM motor drives are proposed to mitigate the overvoltages.

ii
Table of Contents
1. Introduction....................................................................................................................................1
1.1. Motivation .................................................................................................................................1
1.2. Background ...............................................................................................................................3
1.2.1. Surges and overvoltages ...................................................................................................3
1.2.2. Standard Surge waveforms ...............................................................................................7
1.2.3. Metal Oxide Varistors .......................................................................................................9
1.2.4. MOV models .................................................................................................................. 12
1.2.5. MOVs’ failure modes ..................................................................................................... 12
1.2.5.1. Thermal runaway ........................................................................................................ 13
1.2.5.2. Puncture ...................................................................................................................... 14
1.2.5.3. Cracking ..................................................................................................................... 17
1.2.5.4. Flashover .................................................................................................................... 17
1.2.6. Effective Factors in MOVs’ Failure Modes .................................................................... 18
1.3. Problem Statement................................................................................................................... 23
1.4. Dissertation Organization ........................................................................................................ 25
2. Degradation and Health Monitoring of Metal Oxide Varistors ................................................... 27
2.1. Energy Absorption Capability ................................................................................................. 27
2.2. Degradation of MOVs ............................................................................................................. 33
2.3. Degraded MOV’s V-I curve .................................................................................................... 39
2.4. Health Monitoring of MOVs ................................................................................................... 40
2.4.1. Effects of temperature and voltage on health monitoring and lifetime estimation ......... 44
2.4.2. Proposed Health Monitoring Algorithm ......................................................................... 46
2.5. Summary and Conclusions ...................................................................................................... 48
3. Overvoltages in VFDs ................................................................................................................. 49
3.1. Introduction ............................................................................................................................. 49
3.2. Effects of High Frequency Overvoltages................................................................................. 53
3.2.1. Effects on Motor ............................................................................................................. 53
3.2.2. Effects on Drive .............................................................................................................. 55
3.2.3. Other Effects ................................................................................................................... 57
3.3. Causes and Effective Factors ................................................................................................... 58
3.3.1. Voltage Rise Time .......................................................................................................... 60
3.3.2. Cable Length................................................................................................................... 62
3.4. Voltage waveforms in a PWM VFD system ........................................................................... 63
3.5. Simulation of a VFD system for overvoltages......................................................................... 70
3.5.1. Cable model .................................................................................................................... 70

iii
 3.5.2. Measurement of cable parameters .................................................................................. 72
3.5.3. Motor Models ................................................................................................................. 75
3.5.4. Measurement of motor parameters ................................................................................. 77
3.5.5. Experimental setup and Measurements........................................................................... 79
3.5.5.1. Cable parameters ........................................................................................................ 81
3.5.5.2. Motor parameters ....................................................................................................... 84
3.5.6. Simulation results ........................................................................................................... 86
3.6. Summary and conclusion......................................................................................................... 92
4. Overvoltage Mitigation and Protection in VFDs ......................................................................... 93
4.1. Surge Protection in VFDs ........................................................................................................ 93
4.2. Overvoltage Mitigation in VFDs ............................................................................................. 97
4.2.1. Passive Filters ................................................................................................................. 98
a. Filters on Motor Terminals ...................................................................................................... 99
b. Filters on Inverter Output ...................................................................................................... 101
c. Filters on Both Motor Terminals and Inverter Output ........................................................... 104
4.2.2. Active Filters ................................................................................................................ 105
4.2.3. PWM Control ............................................................................................................... 108
4.2.4. DC Cable between Rectifier and Inverter ..................................................................... 112
4.2.5. Energy Varistor ............................................................................................................. 113
5. MOVs in Variable Frequency Drive Systems ............................................................................ 116
5.1. MOV installation on VFD systems........................................................................................ 117
5.2. Behavior of MOV under Repetitive Pulses ........................................................................... 117
5.2.1. MOV’s Response to slow Pulses from a power supply ................................................ 118
5.3. Behavior of MOV under PWM Pulses .................................................................................. 130
5.4. High Frequency Model of MOV ........................................................................................... 135
5.4.1. Simulation of MOV in VFD system ............................................................................. 139
5.5. MOV Application Solutions for VFD systems ...................................................................... 143
5.5.1. MOV in series with GDT.............................................................................................. 143
5.5.1.1. Gas Discharge Tube (GDT)...................................................................................... 144
5.5.1.2. Simulation Results .................................................................................................... 148
5.5.2. TSPD-Switched MOVs (TSMOV) ............................................................................... 150
5.5.2.1. Thyristor Surge Protective Device (TSPD) .............................................................. 152
5.5.2.2. Simulation Results .................................................................................................... 153
5.5.3. Reconfigurable Surge Protective Device (RSPD)......................................................... 156
5.5.3.1. Simulation Results with Ideal Switches ................................................................... 158
5.5.3.2. Switches in RSPD .................................................................................................... 160
5.6. Summary and Conclusions .................................................................................................... 164
6. Conclusions and Future Work ................................................................................................... 165
6.1. Summary of Results .............................................................................................................. 166

iv
 6.2. Suggestions for Future research ............................................................................................ 168
Bibliography .......................................................................................................................................... 173

v
 Chapter 1: Introduction 1

Chapter 1:

1. Introduction

This chapter provides motivation, definitions and background for this research. A literature review on

failure modes of MOVs is presented. Objectives and outcomes of the research are presented.

1.1. Motivation

The continued trend of miniaturization and use of lower voltages in electronics have made electronic

circuits and devices more susceptible to overvoltage transients and surges [1], [2]. Smaller components

with lower energy capacities, closer traces and dense circuits have triggered attention to overvoltage

transients that might not have been seen in the past. Additionally, proliferation of electronic devices in

everyday life and mission critical applications, such as industrial automation, communication, monitoring

and control systems, has made any failure costly due to the repair and service interruption costs. Critical

systems, such as personal computers and communication devices, are being damaged by surges [3]. This

can inflict large financial losses due to downtime, loss of services, and customer dissatisfaction. For

example, a Congressional Research Service study in 2012 estimates the cost of weather-related power

outages at $20 to $55 billion annually, for American businesses and industries [4]. Furthermore, with the

expansion of smart grids and wide collection of data and system information, electronic devices need to be

installed on the power distribution lines and communication towers, where they are exposed to more

frequent and more severe transients and surges [5]. When the increased susceptibility of the modern
 Chapter 1: Introduction 2

computerized systems and their high cost of repair are combined with the fact that the exposure of the

devices to surges has increased, a dire need of a reliable surge protection becomes obvious.

Overvoltage transients and surges are among major failure causes of power electronic converters and

power supplies [6]. Overvoltage spikes damage the electronic devices and electrical appliances in two

primary ways: 1) they may absorb more than the Energy Absorption Capability (EAC) or voltage tolerance

limit of the components. This normally damages the electronic devices instantly and permanently. 2) they

cause extra power loss and heating in the device and degrade it over time. In order to protect against surges

and transient overvoltages Surge Protection Devices (SPDs) are used in electrical appliances and power

systems.

Metal Oxide Varistors (MOVs) are protective devices that are widely used in SPDs and surge arresters.

MOVs are low-cost sintered ceramic blocks with a highly non-linear V-I characteristics. They show high

resistance in normal system voltages and low resistance when an overvoltage incident occurs. An MOV is

placed in parallel with the load and diverts the surge energy away by providing a low impedance path to the

ground for the surge current and maintains the voltage within acceptable limits.

Generally, MOVs are expected to protect the load against infrequent surges and short transient

overvoltages. However, in the power electronic systems, overvoltage transients might be generated within

the system due to the interaction of converters and loads. For example, fast switching of IGBT switches in

Variable Frequency Drive (VFD) systems causes high frequency transient overvoltages that appear on the

terminals of the motor [7], [8]. These repeated stresses can lead to the premature failure of the dielectric of

the motor. Furthermore, these consistent repeating overvoltages pose new challenges to the application of

the MOVs for the surge protection purpose. Frequent high current discharge generates extra heat in the

MOVs that may lead to their thermal runaway and failure. Understanding the surges, transient environment

and the MOVs’ behavior is essential in surge protection of electronic converters.

 Chapter 1: Introduction 3

1.2. Background

1.2.1. Surges and overvoltages

Surges are overvoltage spikes, usually having the lengths of several tens of microseconds. Based on

IEEE definition [9], they do not exceed one-half period of the fundamental frequency in duration.

Although, generally, SPDs are not expected to protect against temporary overvoltages (TOVs),

consideration of TOVs is necessary to ensure the compatibility of SPDs to TOV-related stresses. A

simplified curve of transient environment is shown in Figure 1.1 [9]. It should be noted that the boundary

between No Effect, Upset, and Damage regions are not exact and vary with the withstand characteristics of

the equipment exposed to the surges. The sensitivity of equipment is a key component in evaluating the

effect of voltage sags and interruptions at a facility. However, it is not common to easily obtain the

sensitivity data for components in their datasheets. A curve, developed by CBEMA (Computer and

Business Equipment Manufacturers Association) and revised by ITIC (The Information Technology

Industry Council) [10], describes an AC input voltage envelope which typically can be tolerated (with no

interruption in function) by most information technology equipment. Figure 1.2 shows the ITIC curve.

Figure 1-1. Voltage and duration of transients and their effects on equipment [9]
 Chapter 1: Introduction 4

Figure 1-2. ITIC (CBEMA) curve: Ride through capability of IT Equipment [10]

Mainly, surges are generated by two major sources: lightning and switching. Lightning surges originate

from a direct flash to the power system, or an inductive coupling of a nearby flash into the power line.

Lightning flash density varies from 0 to 14 flashes/sq. km/year in the United States, while in tropical areas

this number can reach 70 flashes/sq. km/year [11]. A cloud-to-ground (CG) flash is typically composed of a

sequence of individual cloud-to-ground return strokes which transfer significant charge from the cloud to

ground, each stroke exhibiting peak currents in the range of 5 kA to 300 kA. These strokes have a nominal

duration of 20-50 microseconds, and are typically separated in time by 20 to 100 msec. A flash will

typically be comprised of 2-3 strokes, but may contain as few as one and as many as twenty strokes [11].

Flash density map of the United States and stroke density map of the world are given in the Figures 1.3 and

1.4. A lightning flash can discharge a huge current, with a typical value of 20 kA, into network, which

causes a severe transient overvoltage. Inductive coupling of lightning can cause severe overvoltages, too. A

one kilometer distant lightning strike can induce about 200 volts in 1 meter of wire [12].

Switching surges are related to the release of the trapped energy in the inductance or capacitance of the

system. They are caused by any switching operations in the power system such as: Load switching,

renewable source switching, or fault clearing. The sudden change in the system can initiate damped

oscillations with high frequencies that depend on resonant frequencies of the network. Magnitude and

duration of the switching overvoltages depends on many parameters such as type of the load and the

switching device [9].

 Chapter 1: Introduction 5

Figure 1-3. US flash density map from National Lightning Detection Network [11]
(Reprinted with permission)

Figure 1-4. World stroke density map from National Lightning Detection Network [11]
(Reprinted with permission)

Overvoltages generated by the switching in power electronic converters are similar to the switching

overvoltages in nature, with the difference that they occur continuously and with high frequencies. Figure

1.5 shows a simple circuit where a simple series RLC circuit is being switched on by an ideal switch at t =

0.005 s. The frequency of the switching transient is determined by values of inductance, capacitance, and

resistance. The peak voltage of the switching transient is dependent on the values of circuit elements and on

the voltage of the power system at the time of occurrence.

 Chapter 1: Introduction 6

Switch turns on

Figure 1-5. Switching transient in a series RLC circuit with ac source

Switch turns on

Figure 1-6. Switching transient in a series RLC circuit with dc source

In most cases, the maximum overvoltage is in the order of twice the peak amplitude of the system

voltage, but higher values can occur, especially when switching on or off inductive loads (motors,

transformers) or capacitive loads. Also, interruption of short-circuit currents can cause high overvoltages

[9]. Figure 1.6 shows the transient response of the same circuit with a dc source. As it is seen from the

figure, the frequency of the oscillations does not change as it is dependent on the values of resistance,

inductance and capacitance of the circuit.

In addition to switching, fault clearing can cause high overvoltages. During a short circuit, a large

current flows from faulty line to ground or to the other lines. Sudden interruption of this large current

creates an overvoltage on lines inductance. Figure 1.7 shows an example of an overvoltage created due to

clearing a fault.
 Chapter 1: Introduction 7

Switch turns on

Figure 1-7. Overvoltage created due to fault clearing

TOVs are defined as power frequency overvoltages that occur in the power system and should be

distinguished from switching overvoltages. For surges, the maximum duration is one half-cycle of the

applicable power frequency [9]. TOVs, on the other hand, occur at the power system frequency. Swells,

overvoltages longer in duration than a surge, but lasting only a few seconds, are considered to be a subset

of TOVs. Typically, existing protective relays and circuit breakers clear the fault within a short time. The

main parameters that influence the amplitude and the duration of TOVs are the type and configuration of

the grounding system and the method of fault clearing in the medium voltage network [9]. The equipment

is designed to tolerate the TOVs in the power system. Generally, SPDs do not have enough energy handling

capability to limit TOVs, but they are exposed to these overvoltages and are expected to ride through these

events with no failure and minimal degradation.

1.2.2. Standard Surge waveforms

IEEE C62.41.2-2002 defines two standard surge-testing waveforms: The Combination Wave, and The

100 kHz Ring Wave [13]. The combination wave is defined by two waveforms: a 1.2/50 µs open-circuit

voltage waveform and an 8/20 µs short-circuit current waveform. The first and second numbers show the

front time and the duration of the waveform, respectively. Figure 1.8 shows the combination wave and its

parameters. The peak value of voltage and current is selected from the standard based on the location

category and specification of the device under test.

 Chapter 1: Introduction 8

(a) (b)

Figure 1-8. (a) Combination Wave 1.2/50 µs open-circuit voltage, (b) Combination Wave 8/20 µs
short-circuit current [13]

IEEE standard defines the front time for voltage waveforms as 1.67 × (t90 - t30), where t90 and t30 are the

times of the 90% and 30% amplitudes on the leading edge of the waveform. The duration is defined as the

time between virtual origin and the 50% amplitude point on the tail. The virtual origin is the point where a

straight line between the 30% and 90% points on the leading edge of the waveform intersects the V = 0

line. For the current waveform, the front time is defined as 1.25 × (t90 - t10), where t90 and t10 are the times

of the 90% and 10% points on the leading edge of the waveform. Similar to the above, the virtual origin is

the time that a straight line between the 10% and 90% amplitude points on the leading edge of the

waveform intersects the I = 0 line. For the Combination Wave, the effective source impedance, the ratio of

Vpeak / Ipeak, is 2.0 Ω.

The Ring Wave is defined as a voltage waveform and no current waveform is specified. It is shown in

Figure 1-9. 100 kHz Ring Wave [13]. The amplitude will decay so that the ratio of adjacent peaks of

opposite polarity is as follows:

- The ratio of the second peak to the first peak is between 40% and 90%.

- The ratio of the third peak to the second peak and the ratio of the fourth peak to the third peak are

between 40% and 80%.

- There is no requirement set on the amplitude of the Ring Wave beyond the fourth peak.
 Chapter 1: Introduction 9

Figure 1-9. 100 kHz Ring Wave [13]

1.2.3. Metal Oxide Varistors

Metal Oxide Varistors (MOVs) are highly nonlinear components that are used to protect the sensitive

circuits against surges and transient overvoltages. An MOV is placed in parallel with the load and diverts

the surge energy away by providing a low-impedance path to the ground for the surge current and

maintains the voltage within acceptable limits. They show very high resistance when voltage is low, around

working voltages, and very low resistance when over-voltages occur. Figure 1.10 shows the basic principle

of MOV application and its V-I characteristic.

Load
MOV

Figure 1-10. MOV Application and Voltage-Current Characteristics

This curve can be divided into three regions. In voltages below the Maximum Continuous Operating

Voltage (MCOV) a small leakage current flows through the MOV and the resistance is high. This region
 Chapter 1: Introduction 10

which sometimes is referred to as linear region shows the operating point of an MOV in normal system

voltage and is called leakage current region. When the system voltage increases above MCOV, the MOV

conducts a current larger than its leakage current and the voltage remains almost constant. This region is

referred to as discharge region or normal operation region. In currents above the nominal discharge current,

the voltage is enough to break all barriers and the only current limiting factor is the grains’ resistance. This

region in which the voltage almost linearly increases with current is called up-turn region.

The voltage clamping characteristic of MOVs results from the complex microstructure of varistor

ceramics. They are polycrystalline materials containing metal oxide grains of random size and shape. Zinc

Oxide is the most common material used in varistors. It is known that back-to-back Schottky barriers form

at the grain boundaries. Switching occurs when the applied voltage exceeds the breakdown voltage, with

the result that the conductivity increases by several orders of magnitude. The non-linear characteristics are

quite dependent on the varistor formulation and the details of the fabrication process.

Figure 1.11 shows an MOV’s simplified microstructure. ZnO grains are highly conductive. At grain

boundaries where ZnO grains are separated by a thin amorphous Bi rich film, back to back Schottky

barriers form. These microvaristors are comparable to the symmetrical back-to-back Zener diodes with an

average breakdown voltage of about 3.5 V. Their series and parallel connections determine the non-linear

behavior of the varistor block.

Figure 1-11. Microstructure of ZnO varistor

Definitions of important specifications of an MOV in this work are as follow.

Maximum Continuous Operating Voltage (MCOV): The maximum designated root mean square (rms)

value of the power frequency voltage that may be continuously applied to an SPD.
 Chapter 1: Introduction 11

Nominal System Voltage: A nominal value assigned to designate a system of a given voltage class.

Clamping Voltage (Also called Let-through Voltage or Limiting Voltage): The maximum magnitude of

voltage, measured at the leads, terminals, or receptacle contacts after the application of an impulse of

specified wave shape and amplitude.

Nominal Varistor Voltage (V1mA): The voltage across the varistor at 1mA, DC.

Voltage Protection Rating (VPR) – A rating selected from a list of preferred values as given in the

standard and assigned to each mode of protection. The value of VPR is determined by the average

measured limiting voltage when applying combination wave of 3kA.

Leakage Current1: The current flowing through the MOV when applying the rated voltage.

Nominal Discharge Current (In): Peak value of the current, selected by the manufacturer, through the

SPD having a current wave shape of 8/20 µs where the SPD remains functional after 15 surges.

Maximum Surge Current: Peak value of the maximum non-repetitive surge current, given by the

manufacturer.

Energy Absorption Capability (EAC)2: The energy that MOV can absorb in one incident of impulse or

TOV, before it fails.

Non-linearity coefficient (α): defined as the slope of the V-I characteristic in normal operation region and

often is calculated as below:

𝑳𝒐𝒈 𝑰𝟐 −𝑳𝒐𝒈 𝑰𝟏
𝜶= Equation 1-1
𝑳𝒐𝒈 𝑽𝟐 −𝑳𝒐𝒈 𝑽𝟏

Usually for MOVs with leakage currents in microamperes, the points 1 and 2 are corresponding to currents

I1=0.1 mA and I2=1 mA on V-I curve.

Time to Failure (TtF)3: time between the application of the overvoltage or surge, until MOV’s

destruction.

1
There is no agreed-upon definition of Leakage Current for low voltage MOVs.
2
There is no standard method to measure EAC. EAC in this work is measured by applying a given voltage until MOV’s
destruction.
3
Not to be mistaken with Mean Time Between Failures (MBTF)
 Chapter 1: Introduction 12

1.2.4. MOV models

A simple equivalent circuit for varistor modeling is shown in Figure 1-12. The main element of the

model is a non-linear resistance, shown as V=f(I). Series resistance and inductance is mainly due to varistor

leads and terminals. Typically, Rs is negligible and Ls is considered only when a surge with a fast rise time

is considered [14]. Values of 𝑅𝑠 = 100 𝜇𝛺 and 𝐿𝑠 = 10 𝑛𝐻 can be assumed for these parameters [15].

Capacitance, CP, can be measured in different frequencies depending on the application.

(a) (b)

Figure 1-12. Varistor Model (a) Simple model (b) IEEE frequency dependent model

The simplest model to represent the non-linear V-I characteristic of an MOV is expressed as 𝐼 = 𝑘. 𝑉 𝛼 ,

where α is the non-linearity coefficient. Based on this assumption, a voltage dependent current source or a

current dependent voltage source can be used to model a varistor. More accurate expressions are also

proposed [16]:

log V = b1 + b2 log (I) + b3 e-log (I) + b4 elog (I) Equation 1-2

1.2.5. MOVs’ failure modes

Although MOVs generally are durable devices, certain mechanical and electrical conditions can lead to

their failure. Most of the times this failure involves high temperatures, and this can lead to fire or damage to

the nearby equipment if no thermal or over current protection is devised. MOVs fail in four different ways:

thermal runaway, puncture, cracking, and flashover [17]–[21].

 Chapter 1: Introduction 13

1.2.5.1. Thermal runaway

Thermal runaway occurs when the generated heat by an increased leakage current rises beyond the heat

dissipation capability of the MOV. This excess of heat would increase the MOV temperature and

consequently, its leakage current. Increased leakage current would generate more heat and would rise the

temperature again. This positive feedback could lead to rapid increase in the MOV’s temperature and a

thermal runaway.

The generated heat, PLoss, due to the Watt loss in the varistor is related to its temperature as

𝐰𝐜
𝐏𝐋𝐨𝐬𝐬 ∝ 𝐞𝐱 𝐩 (− ) Equation 1-3
𝐤𝐓

where wc is the activation energy for conduction, k is the Boltzmann’s constant, and T is the varistor’s

average surface temperature. The dissipated power into the environment is almost a linear function of

varistor’s temperature and can be expressed as

𝐏𝐝𝐢𝐬 = 𝛂 𝐒 (𝐓 − 𝐓𝐚 ) Equation 1-4

where α is the convective exchange coefficient, S is the total heat dissipation surface, and Ta is the ambient

temperature [22]. As shown in Fig. 1.13, PLoss and Pdis may have one or two intersection points. In normal

operating conditions the heat generated in the varistor is equal to the heat dissipated to the environment, so

a balance is kept in point P1. When a varistor absorbs the surge energy, its temperature rises suddenly and

increases power loss generated by the increased leakage current. At such a condition if the varistor

temperature exceeds the stability limit P2 and enters the overheating region, its temperature cannot be

stabilized by heat dissipation and a thermal runaway occurs [21].

Figure 1-13. Thermal stability diagram of an MOV

 Chapter 1: Introduction 14

1.2.5.2. Puncture

Puncture and cracking occur because of imperfect microstructure of the MOV. Non-uniformity of grain

sizes and breakdown voltages of potential barriers at grain boundaries create heterogeneous current

densities within the MOV. This current concentrates into a few narrow paths and heats up those areas more

than adjacent areas and increases the conductivity of those hot spots and paths. This leads to even a higher

current density and increased locally generated heat. The generated heat creates a temperature gradient that

causes thermal stresses between hot spots and neighbor portions, causing MOV block to crack, or increases

the temperature of the hot spots up to the melting point of the ceramic, leading to a puncture failure.

Usually, puncture occurs in low density currents, where there is enough time for the hot spots to form. On

the other hand, cracking failure generally occurs when the current density is high, and stresses become

more than stability strength of the ceramic and crack the MOV before reaching to the melting point. Figure

1.14 shows some MOVs that failed in puncture mode.

Figure 1-14. Puncture failures in MOVs

Varistors are sintered from metal oxide powders which are mixed as well as possible to make the

varistor disks uniform in thermal and electrical characteristics. But always imperfection exists in varistor’s

sintering process and grain growth which makes the threshold voltage to vary in different sections of the

disk’s surface. A measurement by small spot electrodes (3 mm in diameter) in [16] shows this threshold

voltage can vary up to more than 10%. Figure 1-15 shows threshold voltage distribution in 2 different

varistors with 114 mm diameter and 10 mm thickness.

 Chapter 1: Introduction 15

Figure 1-15. Non-uniform threshold voltage distribution in ZnO varistors [23]

This difference comes from the non-uniformity in grain size and threshold voltage of the grain

boundaries. The size of grains depends on the formulation of materials and the sintering process. For the

varistors mentioned above average grain size was 18 μm, and the grain sizes were distributed in the range

of 5-50 μm. The behavior of individual grain boundaries is dependent on varistor’s microstructure. Grain

interfaces containing a thin amorphous Bi-rich film exhibit symmetrical current-voltage characteristic with

a breakdown of 3.6 V in average, where higher threshold voltages are observed, too. Junctions between

ZnO and intergranular Bi2O3 exhibit asymmetrical current-voltage characteristics. Electrons traveling from

a ZnO grain into the intergranular region experience a breakdown voltage at 3.2 V, while breakdown

voltages in the opposite direction are 0.4 and 0.9 volts. Interfaces between ZnO and pyrochlore do not

exhibit distinct varistor properties. [24]

Variation in varistor voltage on the surface of a varistor disk causes the current density to be different in

various portions. Current density is higher in portions with lower varistor voltage and the temperature rises

faster in these portions. Since the leakage current increases with the temperature, this heating increases the

current again leading to “current concentration.” If the current magnitude and duration is enough this

positive feedback process will take place and will concentrate the current in a narrow path through the

MOV. The temperature of that current path will reach the melting point of the ceramic (820 ˚C for Bi 2O3,

1700 ˚C for ZnO) creating a puncture through the MOV and resulting in its failure. The “current

concentration” is a thermal process that takes time. Therefore, it is seen in long duration surges and does

not occur during short duration impulses such as 8/20 µs current surge. Moreover, the longer the impulse is,

the more obvious current concentration will be. The thermal pictures of Figure 1.16 give an evidence of

such facts.
 Chapter 1: Introduction 16

[25]

Figure 1-16. Current concentration in an MOV, gray areas show the highest temperature areas
(Colors changed, originally white); (a) 8/20 µs, 40.8 kA/1700 V, 83-88˚C (b) 10/350 µs, 2.76 kA/920 V,
82-87˚C, (c) 50 Hz, 3 s, 2.04 A/740 V, 110-126 ˚C

Understanding current localization phenomenon helps to interpret some of the MOVs characteristics,

and results of this research on the topic that are presented in Chapter 2. Current concentration increases

with increasing voltage, as the breakdown of some barriers in small areas enables the current to be drawn

into a few paths. But eventually as the voltage is further increased, more and more barriers are overcome,

thus offering more current carrying paths through the microstructure. This delocalization continues until all

the bonds reach the grain conductance, and the system again becomes uniform. [20]

This phenomenon can be better explained by simulation results in [18] shown in Figure 1-17. When

applied voltage is low (700 V/cm) the current distribution is almost uniform. When the voltage increases

the current is concentrated into a few paths. In 2700 V/cm it is obviously seen that current is localized and

most of the current flows from 2 paths. Then increase in the applied voltage can reduce the current

concentration and make current distribution more uniform. As it is seen in 3900 V/cm, there are many

carrying current paths in the varistor block.

Figure 1-17. Current delocalization phenomenon in simulation [18]

 Chapter 1: Introduction 17

1.2.5.3. Cracking

Cracking, like puncture, is caused by high current concentration which results from varistor’s

microstructural non-uniformity and heterogeneous power dissipation and heating in different portions. High

amplitude short duration current surges can quickly inject a high amount of energy into a varistor block. In

a short time of an impulse surge, energy diffusion does not happen inside the varistor, and the heating

process can be considered adiabatic. Therefore, a thermally insulated temperature rise in a part of the

varistor creates temperature gradients between different portions inside the varistor. Different rates of

expansion in different parts of the varistor create thermal stresses on the grain boundaries between ZnO

grains. If these thermal stresses inside the varistor exceed the stability strength of the ceramic, cracking

destruction happens. The critical thermal stress (fc) which causes varistor cracking is in the range between

17.2 and 48.3 MPa. Figure 1.18 shows examples of varistor disks, failed in cracking mode.

Figure 1-18. Cracking failures in MOVs [26], [27]

1.2.5.4. Flashover

Flashover is current concentration on the edge of the MOV, instead of a current flowing through it. In

low-voltage varistors, usually an insulating coating covers the whole body of the MOV and prevents a

flashover. Any failure of the coating due to high temperatures during discharge activities, humidity,

pollution, or material defects can lead to the flashover failure in the MOV. It is conventional wisdom that

high 4/10 µs currents will cause surface flashovers (at magnitudes somewhat larger than 65 or 100kA) [28].

It is also shown in the literature that multi-pulse strikes and sustained over-voltages can cause flashover in

very lower currents comparing to single pulse surges that are common in standard tests [28], [29].
 Chapter 1: Introduction 18

It is a well-known fact that most of the lightning strikes have multiple strokes and up to ten strokes are

quite common. The time interval between these strokes are 15 to 150 ms with an average value of 40 ms.

Darveniza and Saha [28] found that most of the distribution arresters tested by multi-pulse flashes at rated

current or above fail by flashover of the varistor blocks which sometimes involved a little of the metal-

oxide material as well. Figure 1.19 shows some MOVs failed by flashovers.

Figure 1-19. Flash over tracks on MOVs [26]

The probable causes of surface flashover of varistors are [29]:

a) plasma created near the edge of metallization;

b) humidity affecting the varistor outer surface;

c) material defects at the outer edges of the varistor;

d) partial discharges produced at the edges and surface possibly affecting the dielectric strength.

1.2.6. Effective Factors in MOVs’ Failure Modes

Depending on the condition of the varistor itself, ambient conditions, system specifications and surge

parameters different failure modes may occur in MOVs. Although there is extensive research on MOV’s

energy handling capability and failure modes [17]–[19], [23], [24], [30]–[36], not all of the effective factors

are studied or well characterized. Limited surge research laboratories and high cost of tests, differences

between standard tests and real-world application conditions, lack of accurate models for lightning and

switching surges, and randomness of MOV’s behavior due to its complicated microstructure may be some

of the reasons that hinder the research in this area. In this section we will review some of the papers
 Chapter 1: Introduction 19

concentrating on the effective factors in failure modes and energy handling capability. The influencing

factors are:

1) Surge Magnitude and Duration

The dependence of the failure mode on magnitude and duration of the current pulses is well studied by

Eda [19]. As it is seen from Figure 1-20, he showed that in short duration high currents the dominant failure

mode is cracking and in long duration low currents is puncture. However, [19] reported that some puncture

failures are caused by short duration high currents and some cracking destruction are seen in long duration

low current tests. Similar results are reported by [18]. Simulation of thermal behavior of the varistor and

solving heat diffusion equations and calculating the highest temperature and produced thermal stresses due

to temperature differences in hot spots confirms and explains these results as well [17], [37].

Figure 1-20. Effects of magnitude and duration of the pulse on 14-mm varistor’s failure mode

2) Microstructural Non-uniformity of the MOV

As mentioned previously, microstructure of an MOV is random and non-uniform. When non-uniformity

is high, it is predictable that current concentration and formation of hot spots and paths inside the varistor

will cause puncture or will crack the varistor due to thermal stresses [18].

Varistor disks with hot spots of low intensities or very small effective area (or diameter) of the hot spot

usually do not crack and can only fail by puncture. When the area or intensity of hot spot increases the
 Chapter 1: Introduction 20

cracking, failure is seen in high currents. For higher non-uniformities, cracking failure is seen at lower

currents, too [32].

3) Varistor Size

In general, for a varistor, current handling capability is proportional to the cross-sectional area and the

energy handling capability is proportional to the volume, because the bigger the varistor is, the more heat it

can absorb or dissipate. On the other hand, when cross-sectional area of the varistor increases, the

probability of non-uniformity in the microstructure increases, too. So, the energy absorption capability per

unit volume decreases accordingly. This can be seen in Figure 1-21 [18] where 50Hz voltage is applied to

varistors of two different sizes with diameters of 32 mm and 52 mm. The results show a dependency on

varistor size when varistor size is increasing EAC per unit volume decreases that and cracking failure

becomes dominant.

Figure 1-21. Energy absorption capability for different varistor sizes [18]

4) Varistor Coating

In metal oxide varistors an external flashover can occur in currents well below the nominal discharge

currents. So, a high resistance insulation collar coating of glass or other insulating material is applied on the

peripheral surface. In a small varistor, an insulating coating covers the whole body as well as portions of

the connecting leads in order to prevent flashover. The thickness of this coating is important in failure

modes and energy absorption capability.

The coating on a varistor, while providing the electrical insulation, can act as a heat insulator too,

preventing the heat from dissipating into the ambient air, thus reducing the energy absorption capability.
 Chapter 1: Introduction 21

On the other hand, if the coating is too thin, it cannot provide enough insulation in higher currents and

flashover takes place. So, the varistor cannot handle high currents, i.e. the energy absorption capability

reduces. There is an optimal thickness for each coating depending on the material and the varistor size and

specifications.

Based on this reasoning, it is predictable that when the coating is thin the dominant failure mode is

flashover. The experimental tests [38] verify this. It is seen from Figure 1-22 that when the coating’s

thickness increases from 100 µm to 340 µm the failure mode changes from surface flashover to other

failure modes involving the ceramic itself. Furthermore, we can see that when there is an increase in the

thickness of the coating, first the energy absorption capability increases but after a point it starts to

decrease.

Figure 1-22. (a) The effect of coating thickness on failure mode (b) Energy absorption capability [38]

5) Varistor Contacts

Electrodes in the MOVs can be a source of failure. The impulse currents of high density can generate a

large amount of heat and melt the material. Puncture at the edge of the electrode is among the most

common failure mechanisms at high currents [25]. Furthermore, different thermal expansion coefficients of

ceramic body and metal termination plate can cause mechanical stress and cracking along the edges of the

metal plate. Two failed MOV whose failures are related to the electrodes are seen in Figure 1-23.
 Chapter 1: Introduction 22

Figure 1-23. Two failures related to the electrodes [25]

The margin between the edge of the electrode and the edge of the varistor is one of the important factors

in the varistor failure [39]. If the electrode extends to the edge of the varistor the current distribution in the

varistor would be uniform, while the dielectric withstand capability of the varistor would decrease

substantially through surface flashovers. Increasing this margin, on the other hand, will improve the

flashover problem. However, spreading the current from the edge of the electrode will cause a high current

density on the edge, and this will lead to high temperature difference between the elements near the

electrode edge and other portions. Figure 1-24(a) shows the equithermal plots and temperature profiles, for

an impulse current with 250 J/cm3 heat dissipation. In Figure 1-24(b) it is seen that when the margin is 2

mm the temperature difference between the area near the electrode edge and other areas are about 250 °C.

The highest temperature is 650 °C right under the edge of the electrode. This high temperature can be

reduced to 440 °C, shown in Figure 1-24(d), by reducing the margin from 2 mm to 0.2 mm.

Figure 1-24. Equithermal plots and temperature profiles, for 250 J/cm3 heat dissipation [39]
 Chapter 1: Introduction 23

Another problem of the electrode that reduces the energy absorption capability is protrusion of the

metallization in axial direction into the ZnO cavities and defects in the edge of the metallization (non-

smooth edge and radial protrusion toward the edge of the varistor) [20], [39]. Radial protrusion toward the

edge, which is common in metal sprayed electrodes, can change the current distribution and build a hot spot

near the edge. Protrusion of metal electrodes into ZnO cavities is a more severe problem that can contribute

to puncture failure.

1.3. Problem Statement

High frequency overvoltages in Variable Frequency Drive (VFD) systems have adverse effects on

motor and converter operation and lifetime. Also, high voltage spikes cause difficulties in SPD installation

and lower surge protection performance. Originally, MOVs are built to be used in power system

frequencies. In normal voltages and low frequencies of 50-60 Hz, an MOV conducts only a small current.

Over its lifetime, an MOV experiences a limited number of large current discharges due to lightning surges

and TOVs. After every surge, it is assumed that the MOV has enough time to cool down to the ambient

temperature and be ready for the next surge. However, in recent years with expansion of power electronics

in the electrical environment and wide application of VFDs and inverters, MOVs are going to be more and

more installed on lines with high frequency PWM voltages. In such networks, MOV may conduct high

capacitive and resistive currents for a short duration in every pulse. Under such repetitive pulses, the MOV

does not have time to cool down after each pulse. Also, lead and wire inductances may have considerable

effects and may not be neglected. In this research, we investigate the MOV behavior in such high

frequencies, analyze and characterize the related problems. Moreover, we develop and evaluate new

approaches to allow MOVs to operate safely on the line for overvoltage protection and at the same time

decrease the overvoltage in VFD systems.

This research will give insight to MOV behavior in high frequencies to SPD design engineers and

facilitate the application of the MOV in VFD systems as a surge protection device. The results of this

research will influence industry sections that require variable speed operation such as HVAC systems.

Application of MOVs will reduce damages due to external surges, while extending lifetime of the system

by overvoltage mitigation.
 Chapter 1: Introduction 24

In accordance to this vision, the main objectives and contributions of this research are as follows:

Objective 1: Develop a method for health-monitoring of Metal Oxide Varistors (MOVs)

A thorough literature review on MOV’s failure modes are conducted and all the influencing factors are

studied and discussed.

A method for health-monitoring of Metal Oxide Varistors (MOVs) is developed. The proposed method

uses resistive leakage current as an indicator parameter to show the status of health of MOV that can be

used to alarm the user when the MOV is no longer reliable. Moreover, this method calculates the estimated

remaining lifetime of the MOV. This gives the user the ability of proactive maintenance and replacement of

the MOVs before they fail. In order to further increase the accuracy, the proposed method, uses MOV

temperature and under/overvoltage conditions information. Furthermore, considering the behavior of

MOVs and in order to reduce the calculation error, the proposed method takes measurement errors into

account.

Substantial tests on degradation of more than 150 samples of low voltage MOV are conducted and

published for the first time. Surprising findings on Energy Absorption Capability and Time to Failure of

MOVs are observed, investigated and analyzed. These findings give insight to degradation of low voltage

MOVs for application engineers and manufactures.

Objective 2: Analyze, model and explain the reasons for overvoltages in Variable Frequency

Drives (VFDs)

First, a literature review on Variable Frequency Drives (VFDs) was performed to identify their possible

surge related problems, and traveling wave related overvoltages. Characterization of high frequency

overvoltages for the VFD system for different cable lengths, and different pulse rise times are described

and modeled.

An experimental VFD setup, including a rectifier/inverter, a cable, and a three-phase motor, is built in

the lab. High frequency models for cables and motors are studied. Based on the models, impedances of the
 Chapter 1: Introduction 25

existing cable and motor are measured and their model parameters are extracted. The models are simulated,

and results are verified by the experimental measurements.

Objective 3: Develop and evaluate a new method for surge protection in Variable Frequency

Drive systems.

A literature review of overvoltage mitigation techniques, including a comparative study of existing

filters, is conducted. MOV models are presented and examined, and MOV’s behavior is studied under

overvoltages. A model is built for the MOV that is used in this research. The effects of installing MOVs are

analyzed and presented through simulation studies and the measurements on the experimental setup.

Two new surge protection devices are proposed and simulated. These methods reduce the let-through

voltages on the VFD systems due to surges and high overvoltages. The effectiveness of the proposed

devices is verified by simulations and the advantages and disadvantages of the newly proposed methods are

discussed in comparison to the existing method.

1.4. Dissertation Organization

This introductory chapter provides a background on MOVs and explains the definitions and basic

concepts in surge protection. Also, it delivers a literature review on MOV failure modes and factors that

influence these modes. Furthermore, problem statement and objectives of the thesis research are explained.

In Chapter 2, degradation of the MOVs are investigated. Health monitoring methods are studied, and a new

health monitoring and lifetime prediction method is proposed. Chapter 3 explores the high frequency

overvoltage phenomenon in VFD systems and describes the experimental setup that is built in the lab, at

Northeastern University. This chapter, also, includes the modeling techniques and measurements that are

used to simulate the VFD system. In Chapter 4, surge protection principals of a VFD system is described.

Also, overvoltage mitigation techniques are reviewed, and advantages and disadvantages of each approach

are explained. In the final chapter, Chapter 5, MOVs behavior is investigated under pulses and PWM

systems and it is explained that how overvoltages prevent an effective surge protection to be installed in the

VFD systems. Two new solutions to the problem of application of MOVs in VFD systems are developed,
 Chapter 1: Introduction 26

simulated, and discussed in detail. Effectiveness of the proposed methods are demonstrated in comparison

to the existing solutions.

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 27

Chapter 2:

2. Degradation and Health Monitoring

of Metal Oxide Varistors

In this chapter, a literature review on Energy Absorption Capability (EAC) of MOVs is presented.

Then, two experimentally conducted tests on new and degraded low voltage MOVs at Mersen in

Newburyport, MA, are described, and their results of EAC and Time to Failure (TtF) are analyzed. The

research represents the first study on EAC of degraded low voltage MOVs that has been published publicly

[40]–[43]. EAC in AC and impulse currents are measured for the MOVs, and the results are compared with

high voltage station-class MOV results (because there is no other data available on low voltage MOVs).

Furthermore, MOV health monitoring techniques are investigated and effects of temperature and voltage

are studied on health identification parameters. Finally, a method for health monitoring of MOVs is

proposed.

2.1. Energy Absorption Capability

EAC is defined as the amount of energy that an MOV can absorb in an incident of a surge or a TOV

before it fails. This failure can be considered MOV’s mechanical destruction [17], [18], [30] or a

predefined change in its electrical parameters [44]. It is suggested in the literature that EAC is increasing

with increasing current density [17], [30]. Figure 2-1 shows EAC of station-class varistors vs. test current,
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 28

measured by experimental results in [30] and verified by simulation in [17]. It is seen that average EAC

increases from 500 J/cm3 in AC currents around 1 A to above 1500 J/cm3 for pulse currents around 35 kA.

Figure 2-1. EAC of station class varistor disks vs. test current [17]

These experimental tests are conducted on 241 station-class high voltage MOVs, 23 mm high and 63

mm in diameter, with voltage rating of 3 kV [30]. Simulations yielded similar results on station-class

MOVs, as well as distribution-class MOVs of 45 mm high and 32 mm in diameter [17], [37].

Some explanations are presented in the literature to explain the increase of EAC with increase of

current density. It is attributed by [45] to voltage distribution on ZnO grains and grain boundaries. At lower

currents, most of the voltage drop appears on grain boundaries. Thus, most of the energy loss occurs on

small regions of the microstructure. This non-uniform heating creates a temperature gradient and thermal

stress inside the microstructure that causes failure. At high surge currents, on the other hand, the voltage

drops increase on ZnO grains and energy loss and heating are more uniform throughout the microstructure.

However, simulations by [37] reject these explanations by showing that heat transfer at the grain size scale

is too fast to allow any noticeable temperature difference between a grain and its boundaries. They

conclude that increase of EAC at high current densities is a consequence of de-localization of current in the

varistor. At higher currents, the voltage drop on the varistor is enough to overcome the threshold voltage of

boundaries. Thus, current distribution is dependent on the grain resistances. This leads to more uniform

current densities and less current localization. This is also shown by simulations in [18]. Current
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 29

distribution is more uniform in low currents (low voltage drop E=700 V/cm). Current localization increases

to some point, when current increases. Then, at higher currents more current passes are created and current

distribution is more uniform.

Another phenomenon that has received less attention is that, at low currents, initially EAC decreases

with increase in current density [37]. This is again justifiable based on current localization as it is described

in chapter 1.

(a) [18] (b) [37]

Figure 2-2. Energy Absorption Capability vs. current density (a) experimental results (b) simulation
results for different failure modes

Experimental results in [18] show higher EAC at very low currents as well. Figure 2-2(a) shows EAC

decreases from 550 J/cm3 at current density of 2 A/cm2 to around 250 J/cm3 at 10 A/cm2. Furthermore,

simulations in [17] verify these results. Simulation results in Figure 2-2(b) shows the energy needed for any

failure in a distribution-class arrester. It is seen that EAC is higher for low current densities.

In order to fulfill the objectives of the research, we conducted a large number of EAC tests both on new

and degraded MOVs. More than 150 low voltage MOVs are used in the experiment. The specifications of

the MOVs are summarized in Table 2.1. Figure 2-3 shows the size of test MOVs in comparison to the

station-class MOVs used in the references. All the tests are conducted in a UL-certified lab at Mersen USA.

The EAC on the new MOVs are conducted at low density ac currents as well as high density impulse surge

currents. For the ac current test, a constant overvoltage of 240 V is applied to twenty MOVs until their

destruction. For the high current densities, single 8/20 µs impulse currents are applied to MOVs.
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 30

Table 2-1. Specification of Test Specimens

MOV Voltage [V] Size [mm] Max Surge Current Max Energy [J] Power Dissipation [W]
LS40K150QPK53 150 33.5 x 33.5 x 2.5 2 * 40 kA + 20 * 20 kA 360 1.4

63 mm

23-45 mm
33.5 mm
2.5mm

Figure 2-3. Sizes of the station-class varistors [17, 28] (right) and studied low voltage MOVs (left)

2.1.1. EAC test results on new MOVs

Our test results indicate that EAC of tested low voltage MOV is decreasing when the current density

increases. This is different than previously reported EAC for station-class MOVs. AC EAC test results,

which are summarized in Table 2.2, show that EAC of low voltage MOV is in the same range of those of

station-class MOVs reported in [44]. The average EAC per unit volume for 150-volt MOV is 954 J/cm3

when the average test current density peak is 7.8 A/cm2.

Table 2-2. AC EAC Test Results of New 150V MOVs

Average Max Min Standard Deviation
EAC [J/cm3] 954 1142 691 143.7
RMS Test Current [A/cm2] 3.2 6.8 0.6 1.7
Average Peak Current Density[A/cm2] 7.8 16.0 1.6 3.8
Time to Failure [s] 0.67 2.91 0.18 0.60

Table 2-3. Impulse EAC Test Results of New 150V MOVs

Sample # Surge Current [kA] Energy [J] Result
A00 133 4924 Failed
A0 126 4133 Failed
A1 100 2493 Failed
A2 90.4 1904 Passed
A3 90.4 - Failed
A4 84 1800 Failed
A5 80.8 1577 Passed
A5 80.8 1607 Passed
A7 80.8 1549 Passed
EAC [J] 1578
EAC [J/cm3] 574
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 31

1000
954

EAC [J/cm3]
500 574

7.8 7200
0
1 10 100 1000 10000
Current density [Apeak/cm2]

Figure 2-4. Energy Absorption Capability curve for studied MOV

The Maximum 8/20 µs surge that MOV samples can handle is 80 kA. The average energy absorption

during 80 kA surges is 1578 Joules or 574 J/cm 3, which is approximately one third of the values reported

by references [17], [30], [44] for station-class MOVs. The Impulse EAC test results are summarized in

Table 2.3.

Analysis of the failure modes of the test samples provides more insight into the test results. All of failed

samples are examined to determine the failure modes after the test. The main failure modes in impulse tests

are cracking and coating peeling. In impulses higher than 120 kA, the MOVs crack into several pieces.

Usually, cracking occurs outside of the electrode area and a piece under the electrode does not crack. It

seems that the heat generated in the electrode might be a contributing factor to this failure mode.

Furthermore, electrodes provide mechanical reinforcement for the portion between them, while the portions

outside the electrodes are free to vibrate. In lower surge currents, the ceramic body does not crack, and the

main failure mode is coating peeling. Epoxy coating is peeled off mostly in one side, where a round

electrode lies between coating and the ceramic body. Again, excessive heat, generated in the electrode is

the main culprit. Also, the footprints of surface flashovers are seen in some cases.

In low density AC current tests, the main failure modes are puncture and cracking. In samples with

lower current densities, puncture is the only failure mode. When the test current is low, which is the case in

MOVs with high varistor voltage (V1mA), Time to Failure (TtF) is long enough for current concentration
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 32

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

Figure 2-5. Dominant failure modes at (a) high impulse; (b) low impulse; (c) high ac; (d) low ac
currents

and formation of a puncture failure. Puncture is seen in different areas (under, outside and also on the edge

of the electrode) of the ceramic in different samples. In samples with higher AC currents, cracking is seen

beside puncture. The cracking failure here is different than the cracking caused by impulse current, and in

most cases, it splits the whole ceramic body into two or three pieces.

Non-uniformity of the microstructure of the MOV explains the failure modes. In high impulse surges,

the current is concentrated into a few paths due to non-uniform microstructure, creating hot spots and paths

[31]. Thus, high temperature gradient between grains leads to ceramic cracking. However, in lower

currents, hot spots reach the melting temperature of the ceramic, before it cracks, leaving a hole through the

ceramic body. In surge currents that do not crack the MOV, the rapid heating of the electrode and its

expansion may cause the epoxy coating to crack and leave the MOV vulnerable to surface flashovers.

Since the EAC test is conducted with a constant voltage of 240 volts AC, the current is not controlled.

The current depends on the generator impedance and the V-I characteristics of the MOV. In order to

investigate EAC of MOVs in low ac current densities, 12 new MOVs are subjected to lower overvoltages

based on their varistor voltage. This range of current is important in characterizing the behavior of MOVs

during a TOV. AC EACs of all new MOVs (32 in total) are plotted versus peak current density in Figure

2-6. In lower current densities around 0.7 A/cm2, the average EAC of new MOVs is 1413 J/cm 3. This

finding is in accordance with previous studies as described before. When the current density is low and the

time to failure is long, the locally generated heat in the varistor uniformly conducts to the whole varistor, so

the possibility of formation of hot spots is very low. As a result, the energy needed for destruction (i.e.

EAC) would be high.

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 33

Figure 2-6. EAC of New MOVs

One possible fit for the measured data is as below:

EACNew = Ebase + E0 exp(− I peak ) Equation 2-1

Ebase, E0, and λ are constants. Ipeak is the peak value of the MOV current. For the low-density currents

less than 16 A/cm2, this curve is easy to use and beneficial in study of degradation effect. In this equation,

the constant term, here called the base EAC, is almost equal to the average EAC of the group for current

densities higher than 2 A/cm2. For example, for new MOVs shown in Figure 2-6 EAC would be:

EACNew = 947.6 + 705 exp(− 0.821I peak ) Equation 2-2

2.2. Degradation of MOVs

Metal oxide varistors are known to degrade over time when they experience surges and overvoltage

transients. Generally, the leakage current of the MOV increases with degradation, thus the power

dissipation increases within the varistor. Other parameters of the MOV such as varistor voltage change as

well, with degradation [23], [46]. Degradation in MOVs occurs because of the changes in their

microstructure. A high magnitude current surge can change the electrical properties of Schottky barriers at

grain boundaries. Normally, degraded MOVs show oxygen deficiency in the grain boundaries that lower

the height of barriers and increase the leakage current [47]. Also, Bismuth concentration in small areas of

pulsed varistors is observed [48]. The reduction in average grain size and change in the lattice parameters

have been also reported. Inhomogeneity in grain size and threshold voltages can increase the current

concentrations [47]. These changes can be either asymmetrical (polarized) and change the barrier in one
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 34

direction more than the opposite direction, or symmetrical and change both sides equally [23]. Their

collective effects alter the current-voltage characteristics. It is believed that degradation mostly affects the

leakage current region of the V-I curve [49].

Furthermore, it should be noticed that these changes are stochastic processes and each MOV varies

differently. Thus, it is necessary to use multiple test samples in MOV degradation tests to reduce the

possibility of neglecting a particular behavior that might be seen only in a fraction of each production

batch.

Degradation or aging of MOVs due to surges depends on the number of the surges, current magnitude

discharged by the varistor, the duration of the current and its wave shape, the temperature of the varistor,

the formulation of the varistor, and the polarity of the surge and varistor’s previous polarization.

In order to investigate the effects of surge degradation on EAC and TtF of low voltage MOVs, we

conducted substantial tests on surge degradation of MOVs with different number and polarity of surges.

Two sets of MOVs are prepared for this test. The first set (Set A) consists of four groups as described in

Table 2.4. The MOVs of Groups 1 to 3 in Set A are commercially available (CA) ones with no thermal

disconnect terminal. Group 4 MOVs are Custom Made (CM) MOVs that have the same size and

specifications with one thermal disconnect terminal which is clipped to remain close for all tests. The only

difference between Set A and Set B MOVs in their datasheets is that maximum surge is one 40 kA surge

for CA MOVs and two 40 kA plus twenty 20 kA surges for CM ones. The second set (Set B) is composed

of 6 groups of CM MOVs as described in Table 2.5.

The main findings of these tests are described as follows:

1. Average EAC of MOVs decrease with surge degradation.

Decrease in EAC is different in two sets. In Set A, EAC of degraded MOVs after 6 surges of 8/20 µs

decreases 40% from 350 J/cm3 to 211.9 J/cm3. Considering the worst-case scenario, minimum EAC of

degraded MOVs is 80.2 J/cm3 which shows 66% decrease from the minimum EAC of new MOVs, i.e.

236.2 J/cm3. On the other hand, in Set B, comparing Groups 5, 6 (unipolar degradation) and 10 shows that

average EAC reduces 5.5% for Group 5 and 9% for Group 6, after 20 surges of 40 kA. These changes are
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 35

less than what is seen for CA MOVs. This indicates that the CM MOVs are more reliable and resistant

against surge degradation. Furthermore, for the worst case in Group 6, an EAC reduction of 9.5% is seen.

Table 2-4. Set A Samples and Test Results

Minimum Standard
Set A number of Average Average Peak Current
Degradation (8/20 µs Surge) EAC Deviation
Group samples EAC [J/cm3] Density [A/cm2]
[J/cm3] [J/cm3]
1 11 none 350.0 236.2 236.2 3.6
2 11 10% change in V1mA 345.8 195.6 195.6 2.5
3 10 6 impulse of 40 kA 211.9 80.2 80.2 1.0
4 10 none 1081.0 850.0 850 4.2

Table 2-5. Set B Samples and Test Results

Standard Average Peak
Set B number of Average Minimum
Degradation (8/20 µs Surge) Polarity Deviation Current Density
Group samples EAC [J/cm3] EAC [J/cm3]
[J/cm3] [A/cm2]
5 20 2 * 40 kA + 20 * 20 kA 900 660 161 5.0
Unipolar
6 10 20 * 40 kA 870 626 170 2.6
7 10 2 * 40 kA + 20 * 20 kA 953 701 234 3.3
Bipolar
8 10 20 * 40 kA 1036 807 165 1.2
9 20 none 954 691 144 7.8
10 12 none 1413 1029 326 0.7

Figure 2-7. EAC for Groups 5, 6, 7, 8 and new (9-10) MOVs

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 36

In order to understand the degradation effect better, EAC of Groups 5 and 6 is plotted alongside that of

new MOVs in Figure 2-7. The fitted curves can be expressed as below:

EAC5 = 885.8 + 470 exp(−1.15 I peak ) Equation 2-3

EAC6 = 821.7 + 470 exp(−1.15 I peak ) Equation 2-4

It is noticed that Ebase has reduced to 885.8 for Group 5 and 821.7 for Group 6 from 947.6 for new

MOVs. Then, we can see 6.5% for G5 and 13.3% for G6. These numbers are a little higher than those

acquired from average values. The advantage of using E base is that it eliminates the effect of different test

currents, i.e. if a test has more samples in low current area, the average will show higher values, while E base

will not change noticeably, since it comes from curve fitting. Including such information into the varistor

datasheet by manufacturers, will give more insight into product surge capability, compared to the

commonly used method of defining only maximum surge capability.

The importance of using Ebase becomes more obvious when we look at the bipolar degradation tests. It is

seen from Table 2.5 that bipolar surges increase the EAC of MOVs. Although it is true when considering a

constant overvoltage and different currents, using Ebase in equations (2.5) and (2.6) and curves inFigure 2-7,

we realize that for a given current, EAC decreases 10.6% with twenty 40 kA bipolar surges.

EAC7 = 886.2 + 470 exp(−1.15 I peak ) Equation 2-5

EAC8 = 847.2 + 470 exp(−1.15 I peak ) Equation 2-6

2. Unipolar surges decrease EAC more than bipolar surges.

In degradation with unipolar surges, all surges are applied in one polarity, while in bipolar degradation,

after every surge application, the polarity of surge changes for the next surge. Typically, varistor voltage

decrease and leakage current increase are higher in the opposite direction of a surge. The results in Table

2.5 and equations (2.4) and (2.6) show that unipolar surges decrease EAC more than bipolar surges.
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 37

3. Average EAC of CM MOVs is much higher than CA MOVs.

Comparing the average EAC of Groups A1 and A4 shows that the average EAC of CM MOVs is

almost 3 times that of CA MOVs. Average EAC of CA MOVs is 350 J/cm3 that surprisingly low compared

to average EAC of new CM MOVs or values reported in [17], [30] for station-class MOVs. This shows that

some commercial products might have very similar nominal specifications, while they have a significant

difference in their surge capabilities. Part of this may refer to lack of a standard definition for EAC of low

voltage MOVs.

4. Pulsed MOVs conduct lower currents during a TOV.

It is believed that degradation mostly affects the leakage current area of V-I characteristics. However,

there are mentions of changes in TOV regions, also in [49]. Our test results indicate that while surge

degradation increases the leakage current of the MOV, it decreases the MOV current in the TOV region. It

is seen from Table 2.5 that average peak test current for the new MOVs of Group 9 is 7.8 A/cm2, where it is

5.0 A/cm2 for Group 5 and 2.6 A/cm2 for Group 6. MOVs with higher degradation level conduct less

current when a TOV occurs. This is also true for Groups 7 and 8 with bipolar degradation. Furthermore, the

table shows that bipolar surges decrease the test current more than unipolar surges. This is partly because

the bipolar surges, typically, increase the varistor voltage in both directions while unipolar surges decrease

the varistor voltage at least in the opposite direction of the surge.

The change in the MOV’s test current is attributed to change in the TOV region of its V-I characteristic.

Figure 2-8 illustrates the change of V-I curve due to surge degradation. The change in TOV current affects

the EAC of the MOV as well as its TtF, which is described in the next part.

Figure 2-8. Degradation effect on MOV’s V-I curve

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 38

5. Surge degraded MOVs have longer average Time-to-Failure (TtF) than new MOVs for a given

TOV.

Time to Failure is defined as the time from the application of 240 V overvoltage until the MOV’s

failure. Surprisingly, our test results show that the degraded MOVs have longer average TtF than new ones.

This means that surge degraded MOVs tolerate TOVs longer than new MOVs. This is reflected in Table

2.6. As an example, Group 6 MOVs after experiencing 20 nominal surges have average TtF of 2.24

seconds, which means they tolerate a 240-volt TOV almost 3 times longer than new MOVs. Furthermore,

comparison of Groups 5 and 7 shows that bipolar surges increase the TtF more than unipolar surges.

Again, for a thorough investigation, TtF of different groups are plotted and proper curves are fitted. The

linear dependence between logarithm of current and logarithm of TtF of new MOVs is mentioned in the

literature [17, 28, 38]. Figure 2-9 shows the TtF of Groups 5 and 6 compared to that of new MOVs.

Table 2-6. Time to Failures of Set B

Set B Average Minimum

Standard deviation
Group TtF [sec] TtF [sec]

5 1.15 0.3 1.07

6 2.24 0.66 2.04
7 3.17 0.43 3.33
8 10.64 1.11 9.12
9 0.67 0.18 0.60
10 28.2 13.2 20.1

(a) (b)
Figure 2-9. Time to Failure for degraded and new MOVs
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 39

As it is seen from the Figure 2-9(a), TtF decreases at a given current with degradation. Furthermore,

Figure 2-9(b) shows average TtFs, as well as box plots and outliers for new MOVs of G9 and four

degraded MOVs of Groups 5 to 8. The important point is that for a certain TOV, here 240 V, the average

Ttf for the bipolar degradation is higher than that of unipolar surge degradation.

2.3. Degraded MOV’s V-I curve

A simple and widely used model for the MOVs comprises of a nonlinear resistor in parallel with a

capacitor. Degradation changes the V-I characteristics. However, the changes in the separate regions are

completely different. In leakage current region, degradation increases the resistive leakage current. This has

been reported in the literature [46]. Also, nonlinearity factor, α, decreases by degradation [23]. Our test

results on the clamping voltage of low voltage MOVs show that degradation has little effect on the

clamping voltage in high surge currents. Thus, the up-turn region should be considered unchanged with

degradation. This is in accordance with the observations in [44] and [49]. Considering the changes in the

discharge/TOV region is important when considering the behavior of the MOV in TOV conditions. The

results presented in this document show that the VI characteristic moves toward lower currents for a given

voltage in TOV region. In order to incorporate all the above-mentioned changes in a model, we propose to

use the below curve to fit on the V-I data points of the degraded MOV.

𝒍𝒐𝒈 𝑽 = 𝒃𝟏 + 𝒃𝟐 𝒍𝒐𝒈 𝑰 + 𝒃𝟑 𝒆𝒙𝒑(− 𝒍𝒐𝒈 𝑰) + 𝒃𝟒 𝒆𝒙𝒑( 𝒍𝒐𝒈 𝑰) + 𝒃𝟓 (𝒍𝒐𝒈 𝑰)𝟐 Equation 2-7

Figure 2-10. Fitted curve on degraded MOV V-I characteristic

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 40

When b5 = 0, the proposed curve reduces to the curve in Equation 1-2 for new MOVs. b5(log I)2 allows

the curve to fit better in the TOV region. Using this curve fitting in simulations, allows having more

realistic behavior in TOV condition for degraded MOVs.

2.4. Health Monitoring of MOVs

With the emerging concept of the smart grid and growing attention to the asset management and

lowering the down time of services, monitoring of the equipment seems to be an indispensable need in the

near future. Although the failure of a thermally protected varistor does not disconnect the load, it leaves the

sensitive load un-protected and the next surge or transient can damage the load.

Considering the lifetime of an MOV, which is normally more than ten years, the possibility of aging

and failure is not negligible. Despite the progress in the development of MOVs during past decades, a

reliable assessment of their health has not been proposed [48]–[51]. A number of techniques have been

proposed for online condition monitoring of MOVs. Resistive leakage current monitoring is a common

method [52]. Furthermore, total leakage current [53], third and higher harmonics of the total leakage

current [54] and power loss in the MOV [49] have been proposed for MOV health monitoring purposes.

Temperature monitoring has also been used for online monitoring of high voltage metal-oxide arresters

[55].

Different End-Of-Life (EOL) criteria are used by different companies such as leakage current or power

loss reaching double the initial values [56]. In addition to online monitoring techniques, a number of offline

diagnostic methods have been also proposed for MOV health assessment. Destructive approaches such as

Scanning Electron Microscopy (SEM) [47], and X-ray Diffraction (XRD) [48] and non-destructive

methods such as reference voltage measurement [44], discharge voltage measurement [48], and partial

discharge measurement [57] have been proposed in the literature.

Varistor Voltage (V1mA) basically indicates only one point on the V-I characteristic. It has no particular

electro-physical significance but is commonly used as a practical standard reference by manufacturers and

engineers. Usually, 10% change in V1mA from its initial value is considered as unacceptable [58].
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 41

Leakage current monitoring is a common method in Metal-Oxide (MO) arrester condition monitoring

[59]. In normal system voltages, the leakage current of an MOV is a non-sinusoidal current and mainly

capacitive with a small non-linear resistive component. Figure 2-11(a) shows the leakage current of a new

MOV and its resistive and capacitive components. Although the capacitance of the MOV changes slightly

with degradation [60], the capacitive leakage current is generally considered to remain almost constant

through the MOV lifetime. Resistive leakage current increases with aging. However, since it is small

compared to the capacitive component, the change in total leakage current might be small. Figure 2-11(b)

shows the leakage current of a degraded MOV and its resistive and capacitive components. RMS value of

resistive leakage current is increased from 86 µA to 298 µA after twelve surges of 40 kA. Also, the peak

value of leakage current has increased from 221 µA to 787 µA. Although the resistive leakage current

increased 3.5 times, the rms value of the leakage current has increased 35%.

(a) (b)
Figure 2-11. Leakage current (iL), and its resistive (iR) and capacitive (iC) components

The existing methods of extracting the resistive component of the leakage current are mostly based on

orthogonality of the resistive and capacitive currents, such as harmonic analysis and compensation of the

capacitive current [61]. Other methods like point-on-wave [52] are used to extract the resistive leakage

current without the need for the voltage waveform.

Measurement of resistive leakage current and alternatively the power loss of the varistor, need both

current and voltage waveforms. In order to eliminate the need of voltage measurement, third harmonic of

leakage current can be used as an indirect measure of power loss in the varistor. Typically, changes in third

harmonic of leakage current are similar to resistive current. However, the amplitude of third harmonic is
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 42

low, and its measurement is more susceptible to noises in a disturbed electro-magnetic environment.

Furthermore, the third harmonic of voltage, which is unknown without knowing voltage waveform, can

cause a high third harmonic current and create a huge error in third harmonic measurement [39].

In order to investigate the changes in electrical parameters of MOVs with degradation, varistor voltage

and leakage current of all MOVs of Set B (Table 2.5) are measured, and their third harmonics and power

losses are calculated, and the resistive leakage currents are extracted. The method used for the resistive

components’ extraction is the compensation of capacitive leakage currents for all voltage harmonics [61].

Figure 2-12 shows average reference voltages (V1mA) for Groups 6 and 8 of Set B. Voltages are

normalized. Difference in degradation due to unipolar and bipolar surges is obvious in the figure. After a

huge change due to the first 40 kA surge in Group 6, the second unipolar surge of the same size and same

polarity does not change the characteristics remarkably. However, the second bipolar surge restores V 1mA to

values close to the initial values.

Figure 2-12. V1mA vs. number of surges for different groups

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 43

Figure 2-13. V1mA vs. number of surges for Group 5b

The maximum V1mA reduction seen in the samples of the Group 6 is 16 percent after 20 surges of 40

kA. The minimum reference voltage in Group 4 after 20 surges of 40 kA is 0.934 that shows less than 7

percent reduction. For Groups 5 and 7, it is seen that after initial 40 kA surges, subsequent 20 kA surges

increase V1mA both in bipolar and unipolar surges. In order to compare the effect of the initial surge, Group

5 is divided into two sub-groups. Group 5a receives 40 kA surges first, while Group 5b first is subjected to

20 kA surges and 40 kA surges are applied after all twenty 20 kA surges. The result for the V 1mA of the

Group 5b is seen in Figure 2-13. Comparison of two Groups 5a and 5b shows that when an MOV

experience small surges in the beginning, the total reduction in V 1mA is smaller. However, the forward

voltage change might be higher. While the minimum V 1mA in Group 1a is 0.88, none of the Group 5b

samples have V1mA less than 0.975. These results suggest that it is not accurate to describe the degradation

of an MOV simply by the change in its V1mA from its initial value. Especially, in the case of bipolar

degradation, the change in V1mA is negligible, despite the large number of surges and probable changes in

microstructure and voltage current characteristics. Main observations on the electrical parameters are

summarized below.

1. Varistor voltage decreases with unipolar surges of nominal amplitude (40 kA) in the reverse

polarity.

2. Smaller surges (20 kA) mostly increase the varistor voltage in forward direction.

3. The first surge has higher effect on varistor voltage and resistive leakage current.

4. A surge in reverse polarity restores the varistor voltage back to values close to the initial values.

Thus, bipolar surge degradation does not decrease V 1mA.

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 44

5. Resistive leakage current increases with the first surge. However, it might either increase or

decrease by subsequent surges (in both bipolar and unipolar degradation).

Figure 2-14. IR vs. number of surges for different groups

Resistive leakage current (IR) changes with surges, too. The increase in IR is larger in unipolar surges,

when MOV is degraded by large 40 kA impulse currents, as seen in Figure 2-14 for Group 6. It is also seen

that the resistive leakage current decreases when 20 kA surges are applied after 40 kA ones. This trend is

seen in both unipolar and bipolar surge degradation process.

2.4.1. Effects of temperature and voltage on health monitoring and lifetime estimation

The temperature effect on the leakage current of MOVs is known from early days [22]. Figure 2-15

shows a typical change of V-I curve for different temperatures. Leakage current increases with temperature.

Also, increase in the applied voltage increases the leakage current. When monitoring the leakage current or

any of its components, the effects of temperature and over/under-voltage should be compensated.

Figure 2-15. Temperature effect on V-I curve of an MOV [50]

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 45

250

200

Voltage (V)
150
New, 20C
100 New, 100C
After 80*20kA, 20C
50
After 3*50kA, 20C

0
0.001 0.01 0.1 1 10
DC Current (mA)

Figure 2-16. Measured voltage current curve for a new and a degraded MOV in 20 °C and 100°C

Voltage-Current curve of MOVs are measured in the lab by a Keithley Sourcemeter 2410. Measured

voltage-current curve for a new and degraded MOV is shown in Figure 2-16. It is seen that temperature

effect is similar to degradation effect in leakage current region. In order to measure the effects of

temperature and voltage on the low voltage MOVs, the leakage current of 10 new MOVs are measured in

temperatures -20 °C to 85 °C under voltages of 104 V to 150 V. The results are shown in Figure 2-17 for

new MOVs. These temperature and voltage dependency curves change with degradation. The curves for

the same MOVs are shown in Figure 2-18, after degradation process that increased their leakage current to

about 90 µA in room temperature at nominal voltage. For example, if this is considered the end of life of

the MOVs, the numbers from Figure 2-18 give us the upper limits for leakage current in every voltage and

temperature. Interpolation between initial points and the upper limits can provide a rough estimation of the

remaining lifetime of the MOVs.

 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 46

(a) (b)
Figure 2-17. (a) Temperature, and (b) Voltage dependency of resistive leakage current for a new
MOV

(a) (b)
Figure 2-18. (a) Temperature, and (b) Voltage dependency of resistive leakage current for a
degraded MOV

2.4.2. Proposed Health Monitoring Algorithm

Online health monitoring and lifetime estimation of MOVs provides the opportunity to plan for

proactive maintenance and reduces the system down time and operation cost by limiting the maintenance

activities to only near-to-fail MOVs. Based on our tests and measurements, resistive leakage current is the

best parameter to monitor the health of an MOV. Also, compensation for voltage and temperature is

necessary. This research proposes the algorithm in Figure 2-19 to predict the approximate remaining

lifetime of an MOV.
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 47

Capture leakage current

and voltage waveform

Extract resistive leakage current

Measure voltage Extract Resistive leakage current

and temperature lower and upper limits (Figures 2.17
and 2.18)

Estimate lifetime (Interpolation-

Compare the measured value to the
reference values)

Take the proper action (Output,

Alarm…)

Figure 2-19. Proposed health monitoring algorithm for low voltage MOVs

The proposed health monitoring algorithm needs the MOV’s voltage and current waveforms, as well as

its temperature, to estimate lifetime prediction. Sampling frequency should be high enough to extract the

harmonics of voltage and current up to 5 th harmonics. The higher the order of the harmonics considered in

the algorithm, the more accurate the results will become. Resistive current is extracted as discussed

previously. The minimum and maximum values of the resistive current are determined based on offline

measurements on new and degraded MOVs, similar to Figure 2-17 and Figure 2-18. The upper limit is

decided arbitrarily based on the field data of the failed MOVs. Here, we take the values shown in Figure

2-18 as the threshold values. The temperature of the MOV should be measured at the moment of the current

measurement. The temperature sensor should be installed on or close to the surface of the MOV. The

remaining effective lifetime of the MOV, can be calculated using the proposed algorithm.

For example, assume that the resistive current of an MOV is 200 µA, installed in a system with a

nominal voltage of 150 V. Voltage measurement shows the voltage at the moment is 145 V, and the

MOV’s temperature is 40 °C. The lower and upper limits for these conditions can be interpolated from

Figure 2-17 and Figure 2-18. The lower limit, which is the expected resistive leakage current for a new

MOV, is 97.1 µA. The threshold value can be calculated from the degraded MOV curves. Here, it is 261.3
 Chapter 2: Degradation and Health Monitoring of Metal Oxide Varistors 48

µA. Interpolation of the 200 µA in this range, gives us that more than 60% of the MOV’s lifetime has

passed and only about one third of the lifetime is remaining.

remaining lifetime percentage 100% 37.4% 0%

resistive leakage current 97.1 µA 261.3 µA

200 µA

Figure 2-20. Lifetime calculation of the MOV

2.5. Summary and Conclusions

In this chapter, we presented our experimental test results on EAC and TtF of new and degraded low

voltage MOVs using a large number of samples. This is the first study on EAC of degraded low voltage

MOVs that has been published publicly. Our results have appeared in [40]–[43]. Surprising findings are

presented. In particular, it is found that EAC of low voltage MOVs increases in low current densities. Also,

it is found that TtF of degraded MOVs are longer than that of new MOVs in TOV condition. Degradation

effects are explained on the behavior of the MOV and its parameters. Impulse EAC of the low voltage

MOVs are measured by experiment. It is found that impulse EAC of low voltage MOVs are lower than

their AC EAC, contrary to station-class MOVs reported in the literature. Furthermore, different health

monitoring methods are presented, and their advantage and disadvantages of them are explained. Effects of

temperature and voltage on MOV’s electrical parameters are measured and used for estimation of

remaining lifetime of the MOV. Finally, a lifetime estimation algorithm is proposed.

 Chapter 3: Overvoltages in VFDs 49

Chapter 3:

3. Overvoltages in VFDs

In this chapter, a literature review on voltage wave reflection phenomenon and resulted overvoltages on

motor terminals is presented. The effects of overvoltages on the motor, the drive and electric system are

explained. Mathematic formulas are used to calculate the relation between overvoltages, cable length and

voltage rise time. Cable and motor models are discussed, and simulation results are presented.

3.1. Introduction

Electric motors have been the main provider of mechanical energy in industrial and commercial

applications since their invention in 19th century. Specifically, the fixed speed three-phase squirrel cage

induction motor has been used extensively in industry, because of their rigidity and reliability. The speed of

an induction motor depends on only the number of its stator poles and the frequency of the supplied

voltage. Before the advent of SCR1-based speed control solutions, mechanical methods were used to

change the speed of the motor. These methods were usually bulky and inefficient systems that could not

provide continuous speed control over the desired range. Lack of an easy and reliable method to control the

speed was considered as a major disadvantage for induction motors. After development of AC variable

frequency drives in 1960s and 1970s, induction motors became even more popular and prevalent in

industry [62].

1
Silicon-Controlled Rectifier
 Chapter 3: Overvoltages in VFDs 50

A Variable Frequency Drive (VFD) is a solid-state electronic energy conversion system that controls

the speed and the torque of an AC motor by controlling the amplitude and the frequency of the voltage that

is being supplied to it. Generally, this frequency control is achieved by converting AC voltage to DC by a

rectifier bridge converter, and then converting DC voltage back to AC with the desired frequency and

voltage, using an inverter. Usually there is a middle section, called the DC link, comprised of one or several

capacitors that reduces the DC voltage ripples. Various techniques may be used for DC/AC conversion,

such as Pulse Width Modulation (PWM) and Space Vector Modulation (SVM). In these techniques, a

desired output voltage is generated as the average value of multiple discrete voltage pulses, generated by

means of electronic switches. Variable frequency drives may be categorized by their topology, load torque

characteristics, the switching control method, and their output voltage levels. VFDs may be referred to by a

variety of names such as Adjustable Speed Drives (ASD) and Variable Speed Drives (VSD). Figure 3.1

shows a typical drive system with PWM technique.

Figure 3-1. Typical motor drive system

VFDs are used in applications ranging from small electric appliances to large mine mill drives. In recent

years with the increased attention to power efficiency, VFD usage also has increased dramatically in

HVAC applications for energy saving, soft starting and power factor correction purposes. In variable torque

loads such as fans and pumps, VFDs offer a great opportunity for energy saving. Electric motors systems

typically account for about 70% of manufacturing electricity consumption in the US. From this enormous

amount of energy, about 60% are consumed by fans and pumps [63]. As a general rule for variable torque

loads, their power varies with the cube of the speed, e.g., 20% speed reduction would decrease the power

49%.
 Chapter 3: Overvoltages in VFDs 51

𝝎 𝟑
𝝎𝟐 = 𝟎. 𝟖 𝝎𝟏 => 𝑷𝟐 = ( 𝟐 ) 𝑷𝟏 = 𝟎. 𝟓𝟏𝟐 𝑷𝟏 Equation 3-1
𝝎𝟏

In a typical HVAC system, a maximum air flow that the fan should be able to provide, or 100% of

design air flow, is only needed 2% of the time. In 85% of the time, the required air flow is between 40%

and 85% of the maximum air flow [62]. In other words, 98% of the time, energy consumption can be

reduced using a VFD to control the speed of the motors. This provides a huge possibility to reduce the

operating cost by a reasonable amount of investment. The actual saving and payback period depend on the

specifications of the system and the cost of energy. For example, in a hardwood company in New York,

with an initial investment of $46,000 to replace the motors in a ON/OFF moisture control system with a

VFD system, the payback period was less than a year and the company’s first year saving exceeded

$40,000 [62]. For other systems the payback period may vary, but the flexibility and advantages of VFDs

push all industries to use them in all new applications and also replace most of older inefficient systems.

Alongside all advantages that VFDs have and all new horizons that they open up to engineers, they

bring new challenges and new issues to consider, especially when they are retrofitted into existing motor

systems for energy saving or speed control. In most cases, the first question might be whether the new drive

can be installed using the existing wiring and enclosure. However, in some applications, the PWM voltage

may cause other short-term and long-term problems, such as false over-voltage and over-current trips and

reduced lifetime of the equipment [64]. Some of these problems are associated with the over-voltages that

the switching of power electronic devices creates. We will discuss these issues in detail in this chapter.

Usually inverters use IGBT switches with a typical voltage rise of 600 V in 0.1 µs [65]. This rapid

voltage change, alongside the mismatch between the motor and the cable surge impedances, cause high

frequency transients in a VFD system. Reflection of traveling voltage wave from motor terminals creates

high frequency voltage spikes [7], [66]. Figure 3-2 shows the measured overvoltage oscillations on the

motor terminal on the rising and falling edges of one pulse when a 95 feet cable is connecting the inverter

and the motor. Pulse amplitude at the output of the inverter is 350 V. The peak of the overvoltages on the

motor terminal reaches 630 V, which can create numerous problems for the motor, the cable, the drive and

the nearby susceptible electronic devices.

 Chapter 3: Overvoltages in VFDs 52

Inverter output voltage

Motor terminal voltage

Figure 3-2. Overvoltages created on the motor terminal by the reflected voltage wave

In Section 3.3, it is mathematically proved that if the rise time of the voltage pulse is less than twice the

voltage wave’s travel time between the inverter and the motor, a full reflection will occur, and the motor

terminal voltage will reach almost double the voltage of the DC link. Other conditions, such as weak dc

link, neutral point displacement and braking of the motor, might contribute to overvoltages and increase the

amplitude of the voltage spikes. Overvoltages up to 3 times the voltage of DC link have been reported in

the literature [67]. These overvoltages have adverse effects on the motor, the drive, and the electric system,

which will be discussed later in this chapter. Approaches to eliminate or reduce the overvoltages will be

reviewed in Chapter 4.

Other than negative impacts of the over-voltages on the drive system, high frequency voltage spikes

create new challenges for the overvoltage protection devices. The components of the drive system,

especially sensitive electronics in the control circuit need to be protected against external surges and

transients, as well as, overvoltage spikes generated by the drive. Usually, overvoltage and surge protection

devices are designed for occasional events, where they have enough time to cool afterward. When

overvoltages are repetitive in nature, over-voltage protecting devices might experience overheating, failure

or reduced lifetime. These issues, specifically MOV application as a surge protection device will be

explored in Chapter 5.
 Chapter 3: Overvoltages in VFDs 53

3.2. Effects of High Frequency Overvoltages

Advancements in electronic switch fabrication have increased switching frequency while reducing the

losses. Drives have benefited from these fast switches in the form of higher flexibility and performance. On

the downside, the major problems associated with these drives come from their three inherent features: the

fast rise time of the pulses (or high dv/dt), the pulse repetition frequency, and the overvoltage (overshoots).

In a VFD, usually the output voltage of the inverter is near square wave shaped. However, the voltage wave

reflection phenomenon in a cable can create a high frequency overvoltages at the end of the cable that can

surpass twice the DC link voltage. These high frequency voltage spikes, created by the interaction of the

inverter and the cable, have some unfavorable effects on the components of the system [68], [69].

3.2.1. Effects on Motor

Insulating tapes and materials used in induction motors are traditionally designed for power frequency,

50 or 60 Hz, sinusoidal voltage. High frequency and the fast rise time of the PWM pulses, elevate electrical

and thermal stresses developed in the insulators, which can lead to a premature insulation failure. Other

than the high frequency and high dv/dt, the amplitude of the overvoltages also can damage the insulators.

Usually, insulating materials in general purpose induction machines are not dimensioned to resist such high

overvoltages that are imposed to motor windings in VFD systems with long cables. Partial discharge (PD)

occurrence is pointed out as one of the main damaging factors in motor insulations in VFD systems [70]. A

partial discharge is an electrical discharge or spark that bridges a small portion of the insulation between

two conducting electrodes where the electric field strength exceeds the breakdown strength of that portion

of the insulating material. It is known that partial discharge causes erosion and local heating in insulation

materials. These symptoms are frequency dependent, which translates into a higher repetition rate and

higher dielectric loss when higher frequency voltages are imposed. The process of deterioration can

propagate and increase, until the insulation is no longer able to withstand the electrical stress. Industrial

surveys show that the additional stresses generated in the stator winding insulation are one of the leading

root causes of 30% to 40% of induction motor failures [71]. Also, partial discharges emit energy as

electromagnetic emissions, in the form of radio waves, light and heat, and/or acoustic emissions, in the

audible and ultrasonic ranges [72].

 Chapter 3: Overvoltages in VFDs 54

Another phenomenon that can significantly reduce the lifetime of an induction machine is the bearing

current. Bearing current breaks the insulation in the bearings which consists of thin oil film between stator

and the rotor. It can create flashover that erodes and damages the bearing surface and can lead to bearing

failures within a short time [73]. The occurrence of bearing currents in an induction motor has been known

for decades. The early research mainly concentrates on the bearing voltage generated by the alternating flux

linkage of the motor shaft. In an ideal AC motor, driven by AC sine wave power supply, the sum of the

three-phase voltage is zero, thus there is no common mode voltage. However, in a real AC machine three

stator coils are not exactly alike. Asymmetric flux distribution is the main cause of bearing currents inside

the machine. Bearing current in ac power frequency motors has been significantly mitigated by advanced

motor design and manufacturing techniques and no longer is a crucial issue in most applications. However,

recently the problem is again reported when PWM VFDs were installed [74].

The bearing current is caused by the bearing voltage, a voltage generated between the neutral point and

the iron core of the motor, also called common mode (CM) voltage. Unlike a sine voltage system, in a

PWM voltage, the peaks of output CM voltage are high due to the inherent instantaneous imbalance of

three phase voltages. This high amplitude CM voltage in motors in a VFD system alongside with a high

dv/dt creates large pulses of capacitively coupled leakage current that flow between winding and the iron

core of the motor through the bearings. High bearing currents in VFDs increase the electrical erosion and

wear of the bearings, finally resulting into premature failure of the bearings [73].

To understand bearing currents better, three different mechanisms can be identified that cause damage

to the machine’s bearings [73].

1) The CM voltage amplitude can reach to a level that the voltage on the bearing capacitance exceeds

the threshold voltage of the lubricant. With every breakdown, a short-time current flows through

the dielectric and the stored energy in the bearing capacitances is discharged instantaneously. The

electric arc leads to small fusion craters on the inner and outer bearing race that is called electric

discharge machining (EDM). In practical applications, this form of bearing current is associated

with low-to-medium frequency harmonics of the CM voltage.

 Chapter 3: Overvoltages in VFDs 55

2) High frequency ground currents are capacitive currents that are flowing from the motor frame to

rotor ground through the bearings and mainly caused by the CM voltage slope (dv/dt) applied to the

machine terminals. Capacitive bearing currents are dominantly driven by medium-to-high

frequencies of CM voltage.

3) Circulating bearing currents are another capacitive current flowing through the shaft, the two

bearings on the drive end (DE) and the non-drive end (NDE) of the shaft and the stator frame.

Figure 3.2 shows the circulating bearing current mechanism. Both stator and rotor are conductive,

the induced shaft voltage vshaft appears at the two bearings. If vshaft exceeds the dielectric strength of

the insulators in its path, the lubricant in the bearings will be punctured and an alternating high-

frequency current flows through the bearings.

Figure 3-3 Circulating bearing current

3.2.2. Effects on Drive

Due to fast voltage changes of PWM pulses in ac drives, large capacitive currents flow through line to

line and line to ground cable capacitances. These current spikes that are superimposed on normal motor

currents flow at every rising and falling edge of pulses and depending on the rate of the voltage change and

the type of the cable, can have large amplitudes. If the peak value of the current fed by the inverter

surpasses its overcurrent protection setting, an overcurrent trip will occur, and the drive will experience

unexpected shut downs [69], [75], [76].

Figure 3-4 shows a cable’s capacitive charging current for a 95 feet cable. The end of the cable is not

connected to any load and is left open, so all the current seen at the inverter output is cable charging

current.
 Chapter 3: Overvoltages in VFDs 56

L-L voltage, end of cable

L-L voltage
inverter output

L-G voltage, end of cable

Cable charging current

Figure 3-4. Cable charging current, measured for a 95-feet open circuit cable

Furthermore, high repetitive peak currents are flowing through the IGBT switches and contribute to

their heating. Excessive heat by pulse currents can lead to the failure of the power module that usually is

sized for the normal current of the motor [76]. Even if the power switches can tolerate the current spikes,

over-heating and repetitive high currents can reduce their lifetime drastically.

For low-power drives where the bus capacitors are small, the common mode current can pump up the

DC bus voltage and cause an overvoltage trip. The mechanism of feeding power into the dc bus can be

explained by looking at the ground current’s path during the stages of switching of inverter’s switches and

the conduction periods of the diodes in the rectifier. The average power that flows through the dc bus is

nonzero due to the ground current, and can consequently, boost the DC bus level. The boost due to the

common mode current is usually small compared to the bus voltage and will reach a stable amount due to

power dissipation in the inverter and the cables. However, large charging current through the cable

capacitance of long cables can act in the same manner and boost the DC bus voltage. Figure 3-5 shows the

cable charging current path through the DC bus. The drive system comprises of a 380 V converter and

inverter that feeds 75 parallel 3 kW sewing machines through cables that are approximately 100 m long.

Figure 3-6 shows a 200 V rise in the measured dc bus voltage that causes an overvoltage trip [77]. The DC
 Chapter 3: Overvoltages in VFDs 57

bus voltage boost issue is more likely to occur when the drive is attempting to stop the motor or reverse its

direction of rotation [76].

Figure 3-5. Circulation of common mode current

Figure 3-6. DC bus voltage rise due to common mode current [77]

3.2.3. Other Effects

As mentioned previously, in a typical inverter-fed motor, the common mode voltage is noticeable. This

common mode voltage creates a common mode current that has some adverse effects. The transients create

high frequency circulation current into the ground and can trigger upstream differential protection and

ground fault relays [78].

Another issue concerning the ac drives are the generated electro-magnetic interferences (EMI) that can

disturb the susceptible electronic devices [75], [78]. The presence of common voltage in the ac drives

generates high frequency oscillatory ground current in every voltage transition. As switching frequency

increases and machine zero-sequence impedances decrease, the common mode voltage causes larger

common mode currents, worsening EMI problems [78]. Injection of high frequency current into ground

generates radiated emissions as well as noises on the ground wire, disturbing nearby electronic devices.
 Chapter 3: Overvoltages in VFDs 58

3.3. Causes and Effective Factors

When the inverter is connected to the motor through a long cable, because of the cable inductance, the stray

capacitance distributed between the cable wires and the fast rise times of the pulse width modulation

(PWM) voltage from the inverter, will cause overvoltages appear at the motor terminals. Figure 3.2 shows

an equivalent circuit for a transmission line or a cable. Using distributed parameters along the cable, we can

derive:

𝑽(𝒙) = 𝑽+ 𝒆𝜸𝒙 + 𝑽− 𝒆−𝜸𝒙 = 𝑽+ 𝒆𝜸𝒙 (𝟏 + 𝜞 𝒆−𝟐𝜸𝒙 ) Equation 3-2

𝑽+ 𝑽− 𝑽+
𝑰(𝒙) = 𝒆𝜸𝒙 − 𝒆−𝜸𝒙 = 𝒆𝜸𝒙 (𝟏 − 𝜞 𝒆−𝟐𝜸𝒙 ) Equation 3-3
𝒁𝟎 𝒁𝟎 𝒁𝟎

where V and I are the voltage and current at point x, respectively. V+ and V- are constants. 𝑉+ 𝑒 𝛾𝑥 is

complex representation of a forward traveling wave or the incident voltage wave. Similarly, 𝑉− 𝑒 −𝛾𝑥 is

backward traveling wave or reflected voltage wave. Propagation coefficient, γ = α + jβ, is a complex

number in which the real part, α, is called attenuation coefficient and the complex part, β, is phase change

coefficient. Z0 is the characteristic impedance or Surge Impedance of the cable. Γ is the Reflection

Coefficient defined as:

𝑽
𝜞 = 𝑽− Equation 3-4
+

Figure 3-7. Equivalent circuit for sections of a transmission line

For a lossless line, which is an acceptable approximation for a real cable, we will have:

𝒁𝟎 = √𝑳⁄𝑪 (Ω) Equation 3-5

𝜸 = 𝜷 = 𝝎√𝑳𝑪 (1/m) Equation 3-6

 Chapter 3: Overvoltages in VFDs 59

L and C are inductance and capacitance of the cable per unit length. Considering a ZL load at the end of

the line and a source with an internal impedance of Z S in the beginning of the line, we can rewrite the

reflection coefficients at the sending (𝛤𝑆 ) and receiving (𝛤𝐿 ) ends of the cable.

𝒁 −𝒁
𝜞𝑳 = 𝒁𝑳 +𝒁𝟎 Equation 3-7
𝑳 𝟎

𝒁 −𝒁
𝜞𝑺 = 𝒁𝑺 +𝒁𝟎 Equation 3-8
𝑺 𝟎

Reflection coefficient at motor terminal is around 0.9 and for low impedance source is around -1. For an

open line, reflection coefficient is 1 and a full reflection would occur. In this case, the voltage at the end of

the line would reach 2 per unit [79]. Figure 3-8 explains the voltage wave reflection phenomenon for a

voltage pulse with a finite dv/dt on a finite length of cable. Figure 3-8(b) top, shows the incident wave

traveling toward the end of the cable after the switch is turned on. When the inverter switch turns on, it

takes a time, τ, for the voltage wave to reach the motor. Figure 3-8(a) middle shows the incident wave

being reflected on arrival at the receiving end of the line. The incident voltage will be reflected as a positive

voltage, with the same amplitude for an open cable, traveling toward beginning of the cable. The reflected

wave plus the incident wave will cause the voltage equal to double the voltage pulse amplitude at the

terminal of the motor. At time (t1 = τ), voltage amplitude becomes almost twice the amplitude of the

incident voltage, because of the reflection phenomenon. After 2τ, the reflected voltage wave will reach the

inverter. Its reflection from the sending end of cable, called second incident voltage, has the negative

amplitude of 𝛤𝑆 times the reflected voltage wave amplitude. Therefore, the inverter output voltage is barely

affected by the reflection. At the receiving end of the cable, voltage remains almost at two times the

incident voltage until the second incident wave (reflected wave from motor again reflects from inverter)

reduces the voltage level to almost 1 per unit. Also, the reflection of the negative second incident is a

negative wave with the same amplitude. Thus, simultaneously the reflection of the second incident wave

reduces voltage amplitude even more to the minimum at t2. This reflection process repeats several times

depending on the reflection coefficients and creates the voltage oscillation at each rising and falling edge of

the voltage pulse.

 Chapter 3: Overvoltages in VFDs 60

(a) (b)
Figure 3-8. (a) Voltage wave reflection in a transmission line (b) Simulated and measured voltage
waveforms [80]

We can calculate the frequency of voltage oscillations, f0, based on this explanation and the wave

theory. ν is the wave travel speed in the cable. l is the cable length.

𝝂 = 𝟏⁄ (m/s) Equation 3-9

√𝑳 𝑪

𝒇𝟎 = 𝟏⁄ (Hz) Equation 3-10

(𝟒 𝒍 √𝑳 𝑪 )

The overvoltage level depends on the modulation technique, cable length and type, motor impedance,

motor’s mode of operation, and inverter output voltage rise time. Here, we briefly explain two important

factors: voltage rise time, and cable length.

3.3.1. Voltage Rise Time

If we consider the theoretical extreme case of infinite slope for voltage rise, once the voltage reaches the

end of an open cable, voltage rises to twice the magnitude of the voltage pulse. There is no way to avoid it.

In reality, however, the voltage rise time is finite and takes some time for IGBT switch voltage to rise. For

rise times longer than twice the wave traveling time over the cable, the slower the rise time, the smaller the

voltage overshoot amplitude.

The voltage amplitude on motor terminals is the sum of forward traveling voltage waves and reflected

waves. The first incident and its reflection have positive amplitudes, so sum of them create a voltage larger

than the original incident voltage. The second incident, which is the reflection of the reflected wave from
 Chapter 3: Overvoltages in VFDs 61

inverter’s terminal, and its reflection have negative amplitudes. Thus, they decrease the amplitude of the

resultant voltage. This occurs after the voltage wave propagate the cable 3 times.

The first incident and consequently the first reflection have the largest amplitudes and consequent

waves have lower amplitudes. The voltage at motor terminals due to the first two forward traveling waves

is as below

𝟎 𝒕<𝝉
𝒕−𝝉
𝑽𝒓𝟏 = {( ) (𝟏 + 𝜞𝑳 ) 𝑽𝒅𝒄 𝝉 < 𝒕 < 𝒕𝒓 + 𝝉 Equation 3-11
𝒕𝒓
(𝟏 + 𝜞𝑳 )𝑽𝒅𝒄 𝒕 > 𝒕𝒓 + 𝝉

𝟎 𝒕 < 𝟑𝝉
𝒕−𝟑𝝉
𝑽𝒓𝟐 = {− ( ) 𝜞𝑺 𝜞𝑳 (𝟏 + 𝜞𝑳 ) 𝑽𝒅𝒄 𝟑𝝉 < 𝒕 < 𝒕𝒓 + 𝟑𝝉 Equation 3-12
𝒕𝒓
− 𝜞𝑺 𝜞𝑳 (𝟏 + 𝜞𝑳 )𝑽𝒅𝒄 𝒕 > 𝒕𝒓 + 𝟑𝝉

where Vr1 is the sum of the first incident voltage and the first reflected wave. Similarly, V r2 is the sum of

second incident wave (reflected from inverter) and its reflection. Notice that it takes τ seconds for the first

incident voltage to reach the motor. Thus, before this time the voltage is still zero. The same reasoning

explains 3τ for Vr2. In order to avoid full reflection, rise time, tr, should be long enough to allow the second

incident reach the motor before voltage reaches its maximum. This means:

𝒕𝒓 + 𝝉 > 𝟑𝝉 → 𝒕𝒓 > 𝟐𝝉 Equation 3-13

If the voltage pulse rise time is shorter than twice the traveling time, τ, a full reflection will occur, and

motor terminal voltage will reach almost double the voltage of the DC link. At any time, t, after the first

reflection, we can calculate the amplitude of the reflected wave as:

𝒕
𝑽𝒓𝒆𝒇𝒍𝒆𝒄𝒕𝒆𝒅 = 𝑽𝒅𝒄 𝚪𝑳 Equation 3-14
𝒕𝒓

Calculating the receiving end voltage at time 2τ and substituting 𝜏 = 𝑙 ⁄𝑣, we can derive:

𝟐𝒍
𝑽𝒓 𝒑𝒆𝒂𝒌 = 𝑽𝒅𝒄 (𝟏 + 𝝂 𝒕 𝜞𝑳 ) Equation 3-15
𝒓

The relative overvoltage can be calculated as:

𝑽𝒓 𝒑𝒆𝒂𝒌 −𝑽𝒅𝒄 𝟐 𝒍 𝜞𝑳
𝑽𝒐𝒗 = = Equation 3-16
𝑽𝒅𝒄 𝝂 𝒕𝒓
 Chapter 3: Overvoltages in VFDs 62

Equation 3-16 shows the dependence of the overvoltage amplitude to the rise time, as well as the cable

length. The smaller the 𝑡𝑟 , the larger the 𝑉𝑜𝑣 . The relation between the rise time and the overvoltage

amplitude is depicted in the Figure 3-9 based on the Equation 3-16 his figure shows the overvoltage at the

end of a 100 ft cable, connected to a motor with the reflection coefficient of 1. A typical wave propagation

speed of 500 ft/us is used.

1.2
Normalized Overvoltage Peak

1.0

0.8

0.6

0.4

0.2

0.0
0 1 2 3 4
Rise Time (us)

Figure 3-9. Overvoltage on a 100-ft cable for different rise times

Motors and cables have a pre-determined level of insulation against overvoltages, given by the

manufacturer. For a specific overvoltage capability, we can determine the minimum rise time of the

inverter to avoid damages to the motor and cable. For example, if the maximum acceptable overvoltage

level is 1.2 pu, for the cable in Figure 3-9 the minimum voltage rise time is 2 µs. Equation 3-16 can be used

to calculate the minimum rise time for a given cable and acceptable overvoltage.

𝟐 𝒍 𝜞𝑳
𝒕𝒓 𝒎𝒊𝒏 = Equation 3-17
𝝂 𝑽𝒐𝒗

3.3.2. Cable Length

Understanding the cable length’s effect on the overvoltages created in an ac drive system is easy

revisiting the traveling wave theory. If the cable size is short, by the time the voltage wave rises, voltage

wave has propagated the cable back and forth several time. Thus, the voltage at the end of cable is the

resultant of several positive and negative incident and reflected waves that nullify the effects of each other.
 Chapter 3: Overvoltages in VFDs 63

When the cable is long, the interval between the incident voltages allow the voltage to rise/fall after the

subsequent incident negates the voltage change and changes it in the opposite direction.

Equation 3-16 shows a direct proportional relation between the length of the cable and the overvoltages.

It is depicted in Figure 3-10, for an inverter with rise time of 0.2 µs and a typical propagation speed of 500

ft/µs. The reflection coefficient of 1 is used for the receiving end of the cable for simplicity.

1.2
Normalized Overvoltage Peak

1.0

0.8

0.6

0.4

0.2

0.0
0 20 40 60 80 100
Cable Length (ft)

Figure 3-10. Overvoltage for an example inverter and motor for different cable length

In practice, manufacturers specify a maximum cable length for different cable types based on their

inverters. Given the above example, if the maximum acceptable overvoltage is 1.2 pu for the system, the

maximum length of the cable should be limited to around 10 ft. It is also seen that for rise time of 0.2 µs,

we expect a full reflection for a 50-ft cable.

From Equation 3-17, the maximum length of the cable for a given inverter and the maximum acceptable

overvoltage can be calculated.

𝒕𝒓 𝝂 𝑽𝒐𝒗
𝒍𝒎𝒂𝒙 = Equation 3-18
𝟐 𝜞𝑳

3.4. Voltage waveforms in a PWM VFD system

Typically, motor drive systems control motor speed by generating the required voltage and frequency in

the form of repetitive pulses. Depending on the motor and the control strategy, different methods are used

to generate voltage pulses. For an induction motor speed control, space vector modulation (SVM) is a
 Chapter 3: Overvoltages in VFDs 64

common method to control PWM. In this method, as described previously, to create the reference voltage,

two pulses of line to line voltages are created in each switching cycle. Width of pulses depend on what

method of pulse generation or what kind of modulation in SVM is used. However, to generate the same rms

voltage with the same dc link voltage and same switching frequency, the pattern of pulses would be similar

in all of them. Pulses are narrower around 0 and pi and wider around pi/2 and 3pi/2, for a sinusoidal voltage

output. Pulses in the second half cycle are mirror of those in the first half cycle across the zero-voltage line.

Pulse amplitudes are the same for all the pulses, neglecting small variations in dc link voltage.

When surge protection devices and MOVs are installed on 3 phase lines, they may be connected

between a line and ground, between a line and neutral, or between lines. For a traditional power system, all

three voltages have sinusoidal waveform and the line voltage is √3 times the phase voltage. Moreover, for

a symmetrical 3-phase system, voltage between the neutral and the ground is zero. That means line to

neutral and line to ground voltages are basically the same. However, for a PWM system this is not the case.

Pulses are created by connecting lines to either positive or negative dc bus. Considering a midpoint-

grounded dc link, voltage of phases would be either +V dc/2 or -Vdc/2. Thus, for the phase voltages, pulses

would be Vdc high and the maximum voltage would be Vdc/2. Similarly, line to line voltages are either Vdc

or -Vdc, depending on what buses two lines are connected to. The pulse height and the maximum voltage,

both are Vdc. Despite the traditional systems with sinusoidal voltages, line to neutral voltages in a PWM

system are not the same as line to ground voltages. Neutral voltage point is moving to +V dc/6 or -Vdc/6

depending on the lines’ connections to the dc buses. Figure 3-11 shows a simplified diagram of an inverter

feeding a motor, with the grounded DC link midpoint. This is a common configuration in 3 phase ac drives.

Figure 3-11(b) shows the same circuit, during one stage of switching pattern, when S1, S3, and S6 are

closed. At this moment, Phases A and B are connected to the positive bus, and phase C is connected to the

negative bus. The equivalent impedance of two parallel impedances of phases A and B is half of the phase

C impedance. So, voltage is divided 2 to 1 on the motor impedances. V AN and VBN are Vdc/3 and VCN is -

2Vdc/3. In this situation, neutral point moves Vdc/6 towards positive dc bus voltage. For neutral to ground

voltage, we can write:

𝑽𝒅𝒄 𝑽𝒅𝒄 𝑽𝒅𝒄

𝑽𝑵𝑮 = − = Equation 3-19
𝟐 𝟑 𝟔
 Chapter 3: Overvoltages in VFDs 65

(a) (b)
Figure 3-11. (a) Midpoint grounded VFD (b) equivalent circuit when S1, S3 and S6 are closed

When the neutral point of the motor is not grounded, its voltage is equal to the common mode (CM)

voltage. CM voltage is the average voltage of the three phases, by the definition. For the Figure 3-11(b) we

can write:

𝑽𝒅𝒄 𝑽𝒅𝒄 𝑽𝒅𝒄

𝑽𝑨 +𝑽𝑩 +𝑽𝑪 𝟐
+ 𝟐 − 𝟐 𝑽𝒅𝒄
𝑽𝑪𝑴 = = = Equation 3-20
𝟑 𝟑 𝟔

Figure 3-12 shows a typical drive with a diode rectifier and a 3-phase inverter, simulated in OrCAD.

DC link voltage is 340 V and its midpoint is grounded. Three resistors are connected as a Y-connected load

for simplicity to study the line to line, line to neutral and line to ground voltages as well as the CM voltage

created on the inverter output. Figure 3-13 shows the CM voltage, which in this case is equal to the voltage

between neutral point and the ground. As it can be seen from the figure, the peak of CM voltage is equal to

Vdc/2 and quite noticeable in the drive system.

APT30G100BN

V Z1 Z3 Z5
R1 R3 R5
1.5m
Rectifier

C6 1 1 1
PWM1 PHA PWM3 PWM5

AC input V PHB V+ V+

1.5m PHC V-

0 C8 Z2 Z4 Z6
R2 R4 R6

1 1 1

V
PWM2 PWM4 PWM6 RLO1 RLO2 RLO3
100k 100k 100k

VNeg

V-

Figure 3-12. VFD system simulated in OrCAD

Chapter 3: Overvoltages in VFDs 66

Figure 3-13. Common-mode voltage in an ac drive

400V

200V

0V

-200V

(a) -400V
0s 5ms 10ms 15ms 20ms 25ms 30ms 35ms
V(PHA,PHB)
Time

400V

200V

0V

-200V

(b) -400V
0s 5ms 10ms 15ms 20ms 25ms 30ms 35ms
V(PHA,RLO3:1)
Time

400V

200V

0V

-200V

(c)
-400V
0s 5ms 10ms 15ms 20ms 25ms 30ms 35ms
V(PHA)
Time

Figure 3-14. VFD system voltages (a) line to line (b) line to neutral (c) line to ground
 Chapter 3: Overvoltages in VFDs 67

It is seen from Figure 3-14 that line to line voltage pulses have amplitude of 340 V on both positive and

negative directions. The maximum voltage for a line to neutral connection is 226 V, while some pulses

have amplitudes of 113 V and others have amplitudes of 226 V. For line to neutral voltages, it is seen that

all pulses are between -170V and +170V with amplitude of 340 V.

In most VFD commercial products, midpoint is grounded and above-mentioned relations between

voltages are true. However, in some VFDs it is possible that the midpoint of the dc link capacitors is not

connected to the ground. Such a configuration does not affect line to line and line to neural voltages. But

line to ground voltages will be affected. Figure 3-15 shows a VFD system without its midpoint grounded.

This configuration saves one capacitor in the drive, with using one capacitor with higher nominal voltage,

instead of two capacitors in series.

On the power system side, in most cases the neutral wire is connected to the ground. In such a system,

the voltage of the dc link’s midpoint will vary regarding to the ground, causing the line to ground voltages

have higher peak value near the peaks of the input sinusoidal voltage. This elevated line to ground voltage

should be considered in installing appliances between lines and the ground. Figure 3-16 shows the line to

line, line to neutral and line to ground voltages of the system alongside the voltage of the midpoint of

capacitors to the ground. It is seen that the line to line and line to neutral voltages are the same as the

waveforms with the dc link’s midpoint grounded. On the contrary, line to ground voltage is different, when

the midpoint is not grounded. This is a result of dc link’s midpoint voltage swing regarding to the ground,

which is shown in Figure 3-16(d).

Figure 3-15. VFD system without a grounded midpoint

Chapter 3: Overvoltages in VFDs 68

400V

200V

0V

-200V

(a)
-400V
0s 5ms 10ms 15ms 20ms
V(PHA,PHB)
Time

400V

200V

0V

-200V

(b)
-400V
0s 5ms 10ms 15ms 20ms
V(PHA,RLO2:1)
Time

400V

200V

0V

-200V

(c)
-400V
0s 5ms 10ms 15ms 20ms
V(PHA)
Time

400V

200V

0V

-200V

(d)
-400V
0s 5ms 10ms 15ms 20ms
V(C6:2,0)
Time

Figure 3-16. VFD system voltages without midpoint grounding (a) line to line (b) line to neutral (c)
line to ground (d) midpoint of dc link capacitors
 Chapter 3: Overvoltages in VFDs 69

Considering analysis and waveforms above, any change in dc link midpoint voltage will be added to the

phase-ground voltage. Figure 3-17 shows a VFD with single phase input that uses a voltage doubler stage

to boost up the dc link voltage to a required level. This is a common practice in single phase input drives to

create the necessary dc link voltage. In most single-phase systems, the neutral is connected to the ground

somewhere in the system. So, any inadvertent input wire swapping for this drive will change the grounding

connection and will result in doubling the line to ground maximum voltage. Figure 3-18 shows the line to

ground voltage of phase A and DC link’s midpoint voltage. Considering the normal function of the system,

line to line and line to neutral voltages will remain unchanged and the inverter and the motor will continue

to work. However, at the same time, devices installed between the lines and the ground, such as SPDs, will

be exposed to much higher maximum voltage that lead to their failure or faster degradation and reduced

lifetime.

Figure 3-17. VFD system with voltage doubler input stage and swapped phase and neutral inputs

400V

200V

0V

-200V

-400V
0s 5ms 10ms 15ms 20ms
V(PHA) V(Vs:+,0)
Time

Figure 3-18. VFD system voltages with voltage doubler input stage and swapped phase and neutral
inputs (red) line to ground (black) midpoint of dc link capacitors
 Chapter 3: Overvoltages in VFDs 70

3.5. Simulation of a VFD system for overvoltages

Simulation study is a main part of every design and decision making in today’s industry. The key

strength of simulation studies is the ability to investigate the feasibility, functionality and performance of

the proposed ideas without incurring significant expenses. Simulations can answer what-if questions

without interfering with the system and can prevent costly mistakes. It also allows investigation of the

corner cases, where it needs a lot of time, workforce and samples to cover statistically. However, these

benefits come with a few downsides. Simulation time and cost is added to the project and the outputs rely

heavily on the inputs provided and the models used.

Numerous papers are published on the simulation of the ac motor drive systems with long cables [7],

[8], [81]–[87]. Main focus of the simulations might be different based on the goals and target system.

However, in general these simulations are comprised of modelling cable, modeling the motor, and system

simulations. Here, cable and motor models are briefly reviewed first. Then measurements of cable and

motor parameters in the lab are presented. Finally, our existing motor and drive are simulated, and the

results are compared with the experimental measurements.

3.5.1. Cable model

Traditionally, four parameters of a cable are considered in the line and cable models: series resistance

and series inductance that form series impedance, and parallel capacitance and parallel conductance that

present the admittance of the line. In reality, these parameters are distributed over the cable length.

However, considering the lumped parameters makes the calculations easier. Figure 3-19 shows the

conventional cable model which is used in most general studies. This model does not consider the

dependencies of the cable parameters to frequency and consider them to be constant. Distributed model is

briefly presented in section 3.3, by equations 3-2 and 3-3.

 Chapter 3: Overvoltages in VFDs 71

Figure 3-19. Conventional cable model for short cables and low frequency

Most of the studies use variations of the conventional model [7], [66], [85]. Dividing the cable into n

cells and using multiple back to back conventional model for each cell can increase the accuracy of the

model. Paper [7] suggests that a 20-cell RLCG model is an acceptable compromise between simulation

complexity and the size of the error. Another research recommends 5 cells per meter [82].

Other papers add more elements to the conventional model to reduce its errors for high frequencies.

Two of proposed improvements are shown in Figure 3-20 [87], [88].

(a) [88] (b) [87]

Figure 3-20. Improvements of conventional model of the cable

In the applications where the accuracy of the cable impedance over a frequency range is needed, more

accurate models are proposed [82], [84]. Usually, other than three main wires, cables contain one or more

ground wires and shielding layers. Considering capacitances among multiple wires and using the models

that can cover skin effects and frequency dependency of the parameters are necessary in the modeling

accuracy. The model shown in Figure 3-21 through Figure 3-23 considers capacitances between three

phases, ground wire and the cable shield. For better accuracy over a large frequency range, RL and RC

ladder networks are used to closely match the series and parallel impedances of the cable.
 Chapter 3: Overvoltages in VFDs 72

Figure 3-21. Shielded cable model

Figure 3-22. RL ladder network for the series impedance of a cable

Figure 3-23. RC ladder network for the shunt admittance of the cable

3.5.2. Measurement of cable parameters

Generally, calculation of the model parameters is accomplished by measuring common mode, and

differential mode impedances and solving the basic equations. The exact measurement method depends on

the complexity of the model, number of the wires in the cable, and whether the cable is shielded or not. In

the cases where the model is complex, computer programs are used to fit the measured impedance curve to

the model’s impedance.

 Chapter 3: Overvoltages in VFDs 73

For the commonly used 4 wire cable, the measurements of the parameters are described in the Figure

3-24. Inductance of the wire, L, the mutual inductance between 2 wires, M, and the coupling coefficient, K,

can be calculated using the basic circuit theory as below.

(a) Common mode (CM) (b) Differential mode (DM)

Figure 3-24. Common Mode and Differential Mode configurations for cable inductance measurement

(a) Common mode (CM) (b) Differential mode (DM)

Figure 3-25. Inductance measurement for cable in CM and DM mode

𝑳 𝑳 𝟑𝑴
𝑳𝑪𝑴 = (𝟏 + 𝟑 𝑲) = + Equation 3-21
𝟒 𝟒 𝟒

𝑳𝑫𝑴 = 𝑳 (𝟏 − 𝑲) = 𝑳 − 𝑴 Equation 3-22

𝟑
𝑳 = 𝑳𝑪𝑴 + 𝑳𝑫𝑴 Equation 3-23
𝟒

𝟏 𝟑
𝑲= (𝑳𝑪𝑴 − 𝑳𝑫𝑴 ) Equation 3-24
𝑳 𝟒

𝟑
𝑴 = 𝑳𝑪𝑴 − 𝑳𝑫𝑴 Equation 3-25
𝟒

Capacitance of the cable is also measured by the measurement of the line to line capacitance, Cc, and

line to ground capacitance, Cg, as it is depicted in Figure 3-26. capacitances between wires of a cable and

formulated in the following equations.

 Chapter 3: Overvoltages in VFDs 74

Figure 3-26. capacitances between wires of a cable

First, measurement is made by connecting 3 phase wires together and measuring the total capacitance to

the ground, where it will be equal to three times the line to ground capacitance, as it is sown in Figure

3-27(a). The capacitance between wires are also measured by connecting all the wires and the shield

together except one wire. Capacitance between one wire and the rest of the conductors is equal to 2 Cc +

Cg, as it is seen in Figure 3-27(b).

(a) Ca (b) Cb

Figure 3-27. Measurement of capacitance of a cable

𝑪𝒂 = 𝟑 𝑪𝒈 Equation 3-26

𝑪𝒃 = 𝟐 𝑪𝒄 + 𝑪𝒈 Equation 3-27

Then, Cc and Cg can be calculated from the above.

𝟏
𝑪𝒈 = 𝑪𝒂 Equation 3-28
𝟑

𝟏 𝟏
𝑪𝒄 = 𝑪𝒃 − 𝑪𝒂 Equation 3-29
𝟐 𝟔
 Chapter 3: Overvoltages in VFDs 75

3.5.3. Motor Models

The effects of the high frequency voltage components introduced by the PWM technique are usually

neglected when the electromechanical performance of the motor is analyzed. These low frequency

behaviors of the motor can be analyzed by d-q model or conventional T-model equivalent circuit, shown in

Figure 3-28,. On the contrary, motor’s common mode and cable’s charging currents are noticeable under

PWM voltage and cannot be neglected. Overvoltages created by the voltage wave reflection cannot be

simulated with acceptable precision with the conventional induction motor model. For these applications,

high frequency models of the motor have been proposed [81]–[83], [86].

Figure 3-28. Conventional equivalent circuit of induction motor

In order to achieve a high accuracy in high frequency modeling, reference [82] proposes a model with a

per phase equivalent circuit comprised of ten capacitor, ten inductance and ten resistor. The values of these

elements are calculated by the computer programs by the means of curve fitting of the circuit impedance to

the measured impedance over a wide frequency range. The proposed equivalent circuit per phase is shown

in Figure 3-29.

Figure 3-29. Per phase equivalent circuit of motor [82]

 Chapter 3: Overvoltages in VFDs 76

Other researchers [81], [83], [86] have proposed simpler models for high frequency model of the

induction motor. Although they may vary in the configuration of the elements, they all consider the

system’s capacitances to the ground, since in high frequencies the leakage current can be noticeable. A few

of proposed models are shown in Figure 3-30.

(a) [81] (b) [83]

(c) [7] (d) [86]

Figure 3-30. High frequency models of induction motor

There are two main capacitances to consider in the high frequency model of an induction motor:

winding to ground capacitance and winding turn to turn capacitance. In the above models, the capacitance

between w and n represents the winding turn to turn capacitance. The winding to ground distributed

capacitance has been represented by two lumped capacitances. The first one is connected between the

phase terminal, w, and the ground, g. The second one is connected between the motor neutral, n, and the

ground, g. Considering equal values for these two capacitances are favored in most models. The phase

resistance (R or Rt) and the phase inductance (Ld or Lt or LM) are the low frequency parameters obtained by

the locked rotor test. The core and eddy losses are represented by a resistor between w and n (Re or R2). To

increase the accuracy of the high frequency damping, some models consider one or two resistors in the

ground path, in series to the ground capacitors.

 Chapter 3: Overvoltages in VFDs 77

High frequency equivalent circuit provides an acceptable model for the motor for voltage wave

reflection simulation and study of overvoltages. However, it cannot be used to study motor’s power system

frequency behavior, where stablished d-q models or symmetric equivalent are used. To resolve this

limitation, some papers have proposed to include both low frequency and high frequency models as a

universal model for cases that need a wide frequency range [7], [81], [86]. Two of these proposed models

are shown in Figure 3-31. One major drawback of combining low frequency and high frequency models is

the long simulation time. In order to have accurate results for a high frequency phenomenon it is necessary

to use a small timestep. At the same time, to observe the low frequency behavior it is necessary to cover

enough time span. Simulating such a long time with such a fine step, increases of the computation time.

Figure 3-31. Universal (high and low freq) model of an induction motor [81], [86]

3.5.4. Measurement of motor parameters

Generally, parameters of the high frequency model of the motor are calculated using a few

measurements of common mode and differential mode impedances. Sometimes for simplicity, these

measurements are made in a specific frequency in which the impedances and errors are acceptable for the

purpose of the simulation. Alternatively, the parameters can be measured over a range of frequencies and

the calculated impedance curve is fitted to the measured impedance. Taking the model in Figure 3-30(a)

into the account, the measurements basics and equations are explained below.
 Chapter 3: Overvoltages in VFDs 78

Figure 3-32. Motor impedance measurement

(a) Zwn (b) Zwg

Figure 3-33. motor impedances (a) winding to neutral (b) winding to ground [81]

Impedances are measured for these two configurations. The impedance Z wn is measured between the

three phase terminals connected together and the motor neutral, with the floating ground terminal. The

impedance Zwg is measured between the three phase terminals connected together and the ground terminal,

with the floating motor neutral. Calculating these impedances from Figure 3-30(a) with the consideration

that in high frequencies R will be negligible compared to L d, delivers the below equations.

𝒔𝑳𝒅
𝒁𝒘𝒏 = 𝑪𝒈 𝑳 Equation 3-30
𝟑 [𝒔𝟐 𝑳𝒅 (𝑪𝒕 + )+𝒔 𝒅 +𝟏]
𝟐 𝑹𝒆

𝑳
𝒔𝟐 𝑳𝒅 (𝑪𝒕 +𝑪𝒈 )+𝒔𝑹𝒅 +𝟏
𝒆
𝒁𝒘𝒈 = 𝑪𝒈 𝑳 Equation 3-31
𝟔𝐬𝑪𝒈 [𝒔𝟐 𝑳𝒅 (𝑪𝒕 + )+𝒔 𝒅 +𝟏]
𝟐 𝑹𝒆

The impedance Zwn has a zero in the origin and a pair of poles with natural frequency of fpwn.

𝟏
𝒇𝒑𝒘𝒏 = 𝑪𝒈
Equation 3-32
𝟐𝝅√𝑳𝒅 (𝑪𝒕 + )
𝟐
 Chapter 3: Overvoltages in VFDs 79

The impedance Zwg has two complex conjugate zeros at frequency fzwg, as well as one pole at origin and

two complex conjugate poles at fzwn that is equal to fpwn.

𝟏
𝒇𝒛𝒘𝒈 = Equation 3-33
𝟐𝝅√𝑳𝒅 (𝑪𝒕 +𝑪𝒈 )

𝒇𝒛𝒘𝒏 = 𝒇𝒑𝒘𝒏 Equation 3-34

By matching the above equation to the impedance measurements shown in Figure 3-33, parameters of

the model are calculated. Cg and Ct can be calculated using the fact that at both low and frequencies the

impedance Zwg is almost purely capacitive.

𝟏
𝑪𝒈 = @ (𝟏 ~ 𝟏𝟎 𝒌𝑯𝒛) Equation 3-35
𝟔(𝟐𝝅𝒇)𝒁𝒘𝒈

Calculation of Ct shows that it is very small and can be neglected. Re can be calculated using the Zwn at

the resonance frequency.

𝑹𝒆 = 𝟑 𝒁𝒘𝒏 @ 𝒇𝒑𝒘𝒏 Equation 3-36

Ld and R (neglected here) are obtained by the commonly used locked rotor test. This value, however,

might be a little different from the high frequency inductance. For more accurate value we can use

Equation 3-33 to derive the inductance.

𝟏
𝑳𝒅 = 𝟐 Equation 3-37
(𝟐𝝅𝒇𝒛𝒘𝒈 ) 𝑪𝒈

3.5.5. Experimental setup and Measurements

In order to achieve the goals of the research, an experimental setup is built in the lab in Northeastern

University. This setup is comprised of an induction motor controlled by a VFD. Motor shaft is coupled to a

dynamometer as a mechanical load. As the controller, two different drives are used in different tests: one

commercial drive, and one programmable motor control development board. Figure 3-34 shows the

experimental setup.
 Chapter 3: Overvoltages in VFDs 80

Figure 3-34. Experimental set up in Northeastern university lab

The commercial drive is WEG CFW10, a 1 hp motor controller, with a 120 V single phase input. Being

a commercial product, it does not allow much control over parameters such as dc link voltage. Switching

frequency is adjustable from 2.5 kHz to 15 kHz. A voltage doubler at the input creates a DC link voltage of

340 V which is not controllable. The devkit board, made by Microchip, uses a dsPIC33 EP256MC506

Microcontroller. In addition to the motor speed, switching frequency and the DC link voltage are

controllable in this drive.

The cable used in this study, Belden, is a shielded 4 conductor cable specially built for VFD. The

specifications of the cable are provided in the below table.

Table 3-1. Specifications of the cables used in the research

Voltage
Cable AWG Conductors Shield and coverage
rating (V)
BELDEN Aluminum Foil-Polyester Tape 100%
14 4 600
Multicore VFD Braid TC - Tinned Copper 85%

The motor used in the study is a 230 V, 60 Hz, 1760 rpm, 3-phase, 1 hp general purpose motor. The

specifications of the motor are seen in Table 3-2. The shaft of the motor is connected to the shaft of a

dynamometer that can provide a mechanical load for the motor.

 Chapter 3: Overvoltages in VFDs 81

Table 3-2. Technical specifications of the 3-phase motor

Power 1 hp
Nameplate Speed 1760 rpm
Voltage 230 V
Full Load Amps 2.8 A
Nominal Efficiency 85.5 %

Figure 3-35. Motor and cable parameter measurement with the precision LCR meter

Impedances of the cable and the motor is measured in the lab, using a Keysight Precision LCR Meter

(model: E4980A/AL). The test frequency range for this meter is 20 Hz to 1 MHz, with an accuracy of

0.01%. Figure 3-35. Motor and cable parameter measurement with the precision LCR metershows the

impedance measurement of the motor and the cable with the precision LCR meter in the lab. As it is seen in

the pictures, coaxial cables and BNC connectors are used to reduce the impact of unwanted electromagnetic

noises. The meter’s software compensates for the measurement cable’s inductance and capacitance, using

the short circuit and open circuit measurements.

3.5.5.1. Cable parameters

The Belden cable’s common mode and differential mode impedances are measured as it is described in

section 3.5.2. The measurement results are seen in Figure 3-36. The self and mutual inductances can be

calculated using equations 3-23 and 3-25. For example, for inductances in 1 MHz, we have:

L = 618.4 + 3/4 × 349.0 = 880.2 nH/m

M = 618.4 – 1/4 × 349.0 = 531.2 nH/m

 Chapter 3: Overvoltages in VFDs 82

2000

1500

LCM (nH/m)
1000

500

0
10 100 1K 10K 100K 1M
Frequency (Hz)

1200
1000
800
LDM (nH/m)

600
400
200
0
10 100 1K 10K 100K 1M
Frequency (Hz)

Figure 3-36. Cable inductance measurement for Belden Cable (top) LCM (bottom) LDM

Capacitance measurements are also done for the Belden cable as described before and line to line

capacitors and line to ground capacitors are calculated using equations 3-28 and 3-29. As it is seen in

Figure 3-37, measurements show that capacitances for the cable are almost constant in the range of

measurement.

140
120
Capacitance (pF/m)

100
80
60 Cg
40 Cc
20
0
10 100 1K 10K 100K 1M

Frequency (Hz)

Figure 3-37. Line to line (Cc) and Line to ground capacitance of Belden Cable
 Chapter 3: Overvoltages in VFDs 83

Series resistance and parallel conductance of the cable are also measured during the inductance and

capacitance measurements. Parallel conductance can be measured by connecting the three lines wires and

measuring the conductance to the ground. The wire series resistance will be three times the measured

resistance, when 3 lines are connected in parallel. The measured parallel resistance of the cable is seen in

Figure 3-38. Parallel resistance of the Belden cable. The series resistance of the wire can be measured when

measuring the common inductance. It is shown in Figure 3-39 for the Belden cable.

6000
Parallel Resistance (MΩ.m)

5000
4000
3000
2000
1000
0
10 100 1K 10K 100K 1M
Frequency (Hz)

Figure 3-38. Parallel resistance of the Belden cable

250

200
Resistance (mΩ/m)

150

100

50

0
10 100 1K 10K 100K 1M
Frequency (Hz)

Figure 3-39. Series resistance of the Belden cable

Although the impedance of connecting wires is accounted for by the precision LCR meter by measuring

the short circuit and open circuit impedances, still noises and parasitic capacitances and inductances
 Chapter 3: Overvoltages in VFDs 84

interfere with the measurements. To reduce the measurement error, the parameters are measured several

times and average values are used. The cable parameters are seen in the below table for the Belden cable.

Table 3-3. Measured cable parameters for Belden cable

Freq Hz 1M 60
L nH/m 350 957
Rs mΩ/m 196 10.2
C pF/m 130 137
Rp KΩ.m 3.5 299,000
Zc Ω 51.9 83.5
τ ns 196 332

3.5.5.2. Motor parameters

For measurement of the motor parameters, the method described in 3.5.4 is used. In this method Zwn and

Zwg of the motor are measured by a precision LCR meter for frequency range of 20 Hz to 1 MHz. The

measured impedances are seen in Figure 3-40 and Figure 3-41.

10,000

1,000
Zwn (Ω)

100

10

1
10 100 1K 10K 100K 1M
Frequency (Hz)

90
45
Zwn Angle
(Degrees)

0
-45 10 100 1K 10K 100K 1M

-90

Figure 3-40. Zwn measurement for the motor

 Chapter 3: Overvoltages in VFDs 85

1.0E+07
1.0E+06
1.0E+05

Zwg (Ω)
1.0E+04
1.0E+03
1.0E+02
1.0E+01
10 100 1K 10K 100K 1M
Frequency (Hz)

0
10 100 1K 10K 100K 1M
Zwg Angle
(degrees)

-45

-90

-135

Figure 3-41. Zwg measurement for the motor

Measurement results can be used to calculate the parameters of the motor model. Poll and zero

frequencies for Zwn and Zwg are determined from the above figures.

fpwn = 145 KHz, fzwg = 105 KHz, fpwg = 180 KHz Equation 3-38

From equations 3-35 through 3-37 we can write:

Re = 3 × Zwn @ 145KHz = 9.6 KΩ Equation 3-39

𝟏
𝑪𝒈 = = 𝟎. 𝟑𝟑 𝒏𝑭 Equation 3-40
𝟔 (𝟐𝝅 ×𝟏𝟎×𝟏𝟎𝟑 ×𝟏𝟎𝟎𝟎)

𝟏
𝑳𝒅 = 𝟐 = 𝟕. 𝟑 𝒎𝑯 Equation 3-41
(𝟐𝝅×𝟏𝟎𝟓×𝟏𝟎𝟑 ) ×𝟎.𝟑𝟑 ×𝟏𝟎−𝟗

R in the model, in Figure 3-42, is the high frequency frame resistance to the ground. It could not be

derived from the measurements due to the limited frequency range, hence considered 100 ohms. In

summary, the model of the motor can be considered as below:

 Chapter 3: Overvoltages in VFDs 86

Figure 3-42. Motor model parameters extracted from measurements

3.5.6. Simulation results

The cable, the motor and the inverter are simulated in Matlab/Simulink. First, only an open circuit cable

is simulated with distributed parameters. The oscillations of voltage at the motor terminals due to one pulse

is compared to the measured waveforms from the lab setup. Then, the motor model is added to the

simulations for completeness.

As it is seen from the captured waveforms in Figure 3-43, overvoltages are created both on the rising

and falling edges. These waveforms are captured for a dc link voltage of 340 V, but the pulse from inverter

output has small oscillations and its peak reaches 350 V. Using a zoom view at the rising edge allows

measuring the rising time approximately. The curvy waveform of the pulse makes the determination of a

precise rise time for simulations difficult. The voltage rises rapidly at first, but then the slope of the voltage

waveform becomes small. The pulse’s rise time is around 150 to 200 ns until it reaches to around 300 volts

and then slowly increases to 340 volts. The period of the oscillations are around 780 to 800 ns, which

translates approximately to 1.25 MHz oscillations. The time between the pulses at the inverter output and

motor terminal is about 200 ns. The peak value of the voltage on the end of the cable reaches 620 to 640

volts, varying for different pulses.

 Chapter 3: Overvoltages in VFDs 87

Inverter output
Inverter output

Motor terminal
Motor terminal

Figure 3-43. Captured voltages on the experimental setup, open cable

For simplicity, in these simulations, the inverter is considered as a controlled voltage source that

generates an output waveform similar to the measured voltage pulse on the inverter output.

Figure 3-44. Cable simulation with distributed petameters in Matlab Simulink

A pulse with 150 ns rising and falling time is created in the simulation. Pulse reaches 300 V in 150 ns

and then with a smaller slope rises to 350 V by 400 ns after the starting point. More detailed pulse can

create a better waveform on motor terminals that is closer to the captured one. However, for the simplicity,

this pulse can demonstrate the concept.

Figure 3-45. Voltage oscillations on rising and falling edge of a pulse

 Chapter 3: Overvoltages in VFDs 88

Figure 3-46. Voltage oscillations on rising edge of a pulse

Simulations with distributed parameters, using the measured cable parameters shows a 644 V peak

value for the voltage peak and a period of 794.5 ns period of oscillations. Although, we should take into

considerations that the peak value is somewhat dependent on the shape of the pulse, in addition to the rise

time. The period of oscillation is solely dependent on the cable parameters which determines the speed of

traveling wave in the cable. Here, it takes about 200 ns (190 to 210 ns in the measured waveforms, 198.6 ns

in the simulation) to travel across a 29 m cable. A simple calculation gives us the wave speed of 29/0.1986

µs that is 146 m/µs. Calculation of theoretical wave speed, according to Equation 3-9, also shows:

𝟏 𝟏
𝝂= = = 𝟏𝟒𝟖. 𝟐𝟓 × 𝟏𝟎𝟔 Equation 3-42
√𝑳𝑪 √𝟑𝟓𝟎×𝟏𝟎−𝟗 ×𝟏𝟑𝟎×𝟏𝟎𝟏𝟐

Adding the motor model to the simulation, as it is seen in Figure 3-47, changes the waveforms slightly.

However, since the motor impedance is large, the overall changes are not substantial. In fact, some

researchers considered motor as an open circuit in high frequency simulations [80].

Figure 3-47. VFD system simulation in Matlab Simulink

 Chapter 3: Overvoltages in VFDs 89

During the measurements, motor is running in no-load condition. A no-load motor draws a low

frequency sinusoidal current from the inverter. However, in the simulation of the high frequency voltage

oscillations, only a high frequency model of the motor is used, and the low frequency current is zero. Still

the high frequency current can be compared to the captured waveform, because the low frequency current

appears as a dc offset in the captured waveform.

Vab
Vag

Vbg

Ia

Figure 3-48. Captured waveforms on motor terminal

Figure 3-49. Simulated waveforms on motor terminal

 Chapter 3: Overvoltages in VFDs 90

Figure 3-48 shows the captured waveforms of line and phase voltages of the motor in the experimental

VFD system, as well as motor’s current. When the cable is connected to the motor, for a 340 V pulse, the

peak value of line to line voltage at the motor terminal reaches 610 to 620 volts. The oscillation period is

200 to 210 ns. Line to ground voltage is varying between 550 to 570 volts for the captured pulses. The

waveform of the line to ground voltage is also different than the line to line voltage, as the voltage of the

other phases do not remain at the dc link level because of the induced voltage. In Figure 3-49 the green

waveform is the motor current with its low frequency value of 1.6 A at the time of the capture. The high

frequency component of the current is super-imposed on the low frequency current and its peak reaches 1.3

amps for the first oscillation in the pulse. This value for the simulation is 1.1 amps. The high error value in

capacitive current measurement might be related in part to the capacitance between the length of cable after

the current probe and the motor frame. It might also be a result of not considering all the parasitic

capacitances of the motor.

Figure 3-49 shows the same waveforms from the simulation. The peak value of the line to line voltage

is 619.6 V, where the peak value of line to ground voltage for phase a is 582.5 V. The peak value of the

induced voltage reaches 182 volts in the simulation, where its measured value was 180 volts. The main

variables are summarized in Table 3-4.

Table 3-4. Comparison of simulation results to measured values

Measurement Simulation Error (%)

VL-L peak (V) 610 619.6 1.5
VL-G peak (V) 200 204.7 2.3
τ (ns) 570 582.5 2.1
iMotor (A) 1.3 1.1 -15.3

When designing a system, or analyzing the system for a particular issue, simulation of the system can

prove useful. Simulation can make impossible scenarios to be studied or possible investigation easier. For

example, calculating the voltage peak vs. cable length is possible to be done on a real system. However, it

needs multiple cables with various length and cables are costly. Moreover, calculating the voltage peak vs.

the voltage rise time might be even more difficult in practice, while in the simulation it can be easily done.
 Chapter 3: Overvoltages in VFDs 91

Simulation studies also provide important details that otherwise usually are neglected in simplified

calculations and approximations. For example, relation between the maximum overvoltage and the rise

time of the pulse can be oversimplified neglecting the effect of the pulse waveform and assuming a linearly

rising voltage pulse. Moreover, motor impedance in the end of the cable also is not an ideal pure resistive

impedance or open circuit to reflect the same waveshape as the incident voltage wave. Calculating the

overvoltage using Equation 3-16 leads to monotonically decreasing overvoltage values as the rise time

increases. It can be seen in Figure 3-9. However, the waveform of the pulse can change this trend.

Overvoltage value can increase for certain values of rise time, if rise time increases. Figure 3-50 shows the

simulation results for the peak value of the voltage on motor terminal versus pulse rise time. As it is seen, if

the rise time is less than 2τ a full reflection occurs. This was expected as it is explained previously.

However, for the rise time values larger than 2τ (0.415 to 0.83 µs), the overvoltage value is less than the

calculated one. This is because of the pulse and reflection waveforms have larger slope in the beginning

and smaller slope near the max value. When the second reflection arrives at the end of cable, it decreases

more than what the last part of the incident wave increases. This leads to a lower overvoltage on motor

terminal.

Another important result is that, not always increasing the rise time can lead to the decreased

overvoltage. In this case, if the rise time is about 0.9 µs and we increase it to 1.3 µs, not only we do not

decrease the overvoltage, but also, we increase from about 10% to 20%.

1
Normalized Overvoltage

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5
rise time (µs)

Figure 3-50. Simulation results: motor's voltage peak vs. pulse rise time
 Chapter 3: Overvoltages in VFDs 92

3.6. Summary and conclusion

In this chapter, the theory of the overvoltages in the VFD systems and their adverse effects on the

components of the system are presented. Also, the effective factors on the amplitude of the overvoltages are

analyzed through theoretical and simulation studies. It is shown that overvoltage values are mainly

dependent on the pulse rise time and the cable length. In summary, if the voltage pulse rise time is shorter

than twice the voltage wave’s traveling time for the cable, τ, a full reflection will occur, and motor terminal

voltage will reach almost double the voltage of the DC link. Furthermore, in this chapter, cable’s and

motor’s high frequency models are discussed, and the measurement techniques of the model parameters are

presented. High frequency models are extracted for the VFD system that is built in the lab in Northeastern

University and the system is simulated. Simulation results are shown to match the captured waveforms and

measured values. It is also revealed that over-simplification of some formulas might lead us to wrong

conclusions, where simulation results can provide a detailed analysis with acceptable accuracy.
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 93

Chapter 4:

4. Overvoltage Mitigation and

Protection in VFDs

In this chapter, surge protection requirements of the ac drive systems and the locations of SPDs in a

VFD system are discussed. Furthermore, regarding the high frequency overvoltages created by the drive

that are discussed in the previous chapter, here, a literature review on the existing approaches to mitigation

of those overvoltages and protection against them in VFD systems are presented. Advantages and

disadvantages of those techniques are discussed.

4.1. Surge Protection in VFDs

As any other electric system, VFD systems need to be protected against surges and overvoltage

transients. These surges can come from the utility side or be generated by the drive itself. Typically, the

surges that come from the power system are less frequent and have higher energy and amplitude. These

surges could be lightning surges, or the switching surges from the power system. In addition to those

surges, operation of the converter/inverter also can create overvoltages that can be detrimental to the

sensitive electronic circuits. An effective surge protection of a drive system should protect the power

electronic switches and control circuit as well as the motor.

 Chapter 4: Overvoltage Mitigation and Protection in VFDs 94

In a typical drive system, there are five points to place surge protection devices, shown in Figure 4-1.

The SPDs used in these locations might use protective devices with different technologies. In the

commercial products, some SPDs might be integrated with other products, such as filters to provide

protection against poor power quality or high harmonic distortion. In these cases, installation of those

hybrid products need extra considerations. From surge protection viewpoint, each location in Figure 4-1 is

discussed below.

Figure 4-1. One-line diagram of a typical drive system and its surge protection

Location 1 - Drive Input: The first and the most important point for protecting the drive system is at

the drive input. Surges from utility can have large energies and/or amplitudes and can damage the rectifier

diodes and inverter switches and even affect inverter output [89], [90]. An SPD at this point provides

protection from events propagated on the electrical system from upstream sources and external events such

as lightning and switching surges created by the utility. An IEEE application note recommends parallel

SPD with no filtering at this point in order to avoid any interaction and interference with drive’s capacitive

components [91].

Location 2 - Control Circuit: The control circuit contains sensitive electronics and can be damaged by

surges from external sources or by the transients created by the drive. In most cases the input voltage of the

drive is 480 V, and the control circuit is isolated by a step-down transformer. It is recommended that a two

port SPD with filtering to be installed at location 2 in order to provide low protection level, needed by the

electronic circuit [91]. In low voltage drives, usually the control circuit and the power circuit and switches
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 95

are integrated into one board. In this case, the protection of the control circuit is recommended by SPDs on

the board.

Locations 3 and 4 - Drive Output and Motor Input: Protecting the immediate drive output is

recommended, when the length of the connecting cable between the drive and the motor is longer than 10

m (30 ft) or if the connection is routed along an external wall or outdoors. An SPD at this point diminishes

the effects of direct lightning or induced voltage surges due to nearby lightning. These surges can cause

damage to the drive, even if protection is provided at the motor input. Protection of the motor input is

essential, too. Providing protection at this location helps extend the life of the motor and prevent damage to

the windings and bearings of the motor due to surges. At these locations, a parallel connected SPD is

appropriate [91].

One important issue at locations 3 and 4 arises from the fact that majority of inverters use Pulse Width

Modulation (PWM) techniques to control the frequency and the rms value of the voltage. Thus, the

maximum of the waveform is equal to the DC link voltage. This maximum voltage is normally higher than

the peak value of a sine waveform with the same rms value (𝑉𝑝𝑒𝑎𝑘 = √2𝑉𝑟𝑚𝑠 ). When the maximum voltage

is considered to select a proper SPD, the protection level would be higher compared to a motor without a

VFD. For example, consider a 208 V three phase induction motor, fed with sinusoidal voltage. Line to Line

voltage has the maximum of 294 V. Line to ground rms voltage is 120 V, with the peak value of 170 V.

Line to neutral voltage is equal to line to ground voltage. For a proper MOV selection, the maximum

voltage at the point of installation is considered. Usually MOVs are marketed with a limited number of

rated MCOVs that engineers select from. For example, one can choose to install 150V MOVs between

lines and ground, and lines and neutral, and 275 V or 320 V MOVs between lines, depending on the study

of overvoltages at that point. This is a common and well-stablished practice and all MOVs are designed and

tested for those conditions. However, in an electric motor equipped with a VFD speed controller, peak

values of line to line, line to neutral and line to ground voltages are dependent on the DC link voltage. For

example, a drive that controls a 208V three phase motor is likely to have a 340 V dc link. In this case, the

maximum line to line voltage would be 340 V, while the line to neutral and the line to ground voltages

would have the peak values of 226 and 170 volts. It forces the engineer to select a higher rated MCOV,

such as 420 V for line to line protection. Choosing a higher MCOV, seems as a reasonable choice because
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 96

it allows the system to function. However, it also increases the voltage levels that the protected

system/appliance is exposed to. For example, for a 5-kA surge that is not rare in a distribution system, the

clamped voltage of 420 V MOV is 400 V higher than that for 320 V MOV [58]. Even if the appliance can

tolerate these higher overvoltages, it can dramatically reduce its lifetime.

Another difficulty in the surge protection for the locations 3 and 4 in the Figure 4-1 is the lack of

application notes and installation guidelines. Typical datasheets of MOV manufacturers do not consider

installations on PWM lines. Almost no documents are available that describe the selection criteria of

MOVs for such lines or that describe the long-term effects of PWM pulses on MOV’s reliability and

lifetime expectancy. Further, typical manufacturer post production and quality check tests are based on the

traditional power system sinusoidal waveforms.

Another important and troublesome issue in installation of MOV-based SPDs on motor terminal is the

existence of high frequency voltage reflection overvoltages, especially when the connecting cable is long.

As discussed before, when there is no limiting device, these voltage spikes can reach the amplitudes of

twice the DC link voltage. MOVs installed on the motor terminal conduct high currents during these peaks,

which can result in MOV failure or an extreme reduction in MOV’s lifetime. A simple solution that

engineers might use is to install an MOV with higher MCOV in order to avoid high leakage current.

Similar to what is mentioned above, this would increase the protection level which might not be acceptable,

depending on the motor insulation level.

Location 5 - Series SPD at Drive Input: At location 5, a series surge protector can add extra security

for the drive’s surge protection. As mentioned above, typically a parallel surge protection is needed at

drive input. Moreover, a line reactor can protect the drive against lightning strikes and voltage spikes.

Reactors may be installed for use on either the drive input or drive output for different purposes such as to

balance the ac lines, or to limit the short circuit current. When installed on the dive input, reactors provide a

level of protection from the incoming surges from the utility side. When reactors are placed on the drive

output, they help reduce the voltage fast rise time (dv/dt) and the high peak voltages which can be

experienced in IGBT inverter applications when the distance between the inverter and the motor is long.

The degree of effectiveness of the reactor depends mostly on its impedance. The most common selections
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 97

for this are 3% and 5% impedance reactors [92]. Figure 4-2 shows the effect of an incoming voltage spike

from AC power lines on DC bus voltage. Voltage spikes on the AC power lines cause elevation of the DC

bus voltage which may cause the inverter to trip for protection. Reactors can absorb these line spikes and

offer protection to the rectifiers and DC bus capacitors while minimizing unnecessary tripping of the

inverter.

Line reactors have several side effects that need consideration. The main side effect is the voltage drop

on the reactor and line loss due to the coil’s resistance. Thus, the use of a reactor is dependent on whether a

DC link choke is built into the drive.

Figure 4-2. Effect of a voltage spike from AC power line on DC bus voltage Mitigation of Transient
Overvoltage in VFDs [92]

4.2. Overvoltage Mitigation in VFDs

The phenomenon of voltage wave reflection at motor terminals in a PWM AC drive creates large high

frequency overvoltages and imposes severe stresses on the motor insulation. If no mitigation measures are

taken, the drive system will face higher damage possibility, longer down time, reduced reliability and

shortened lifetime. When the drive/motor system is small and low-cost, installing extra devices for

mitigation of these overvoltages might not be economical. In these conditions, depending on the economic

assessments, installing a motor and a cable with a higher insulation strength might be an acceptable

decision [93]. For larger drive systems or when other engineering design constraints must be met, a
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 98

mitigation technique might be used, based on the severity of the overvoltages and financial assessments.

Several approaches have been proposed in literature to solve the reflected overvoltage problem in motor

drive systems. Here we summarize the benefits and shortcomings of each approach.

4.2.1. Passive Filters

Passive filters are the most common choice to mitigate the overvoltages on the motor terminal. Passive

filters are used either on motor terminals, inverter output or both. The design goal for the filters on motor

terminals is to match the impedance of the load with the surge impedance of the cable [94]. Matching surge

impedance eliminates the reflection of the voltage at the motor terminal and theoretically reduce the

overvoltages to zero. Similarly, the philosophy behind using the filters on inverter output is to increase the

rise time of the pulses [95]. A slower rise time provides enough time for consecutive reflections to arrive

and reduce the effect of the first reflection. Hence slow-rising pulses create smaller overvoltages. Both

approaches require installation of extra components on the line. Apart from the cost and the need for a

space to be installed, a filter adds to the losses of the system and/or causes a voltage drop. The ultimate

goal is to make an agreement among the conflicting goals: minimal overvoltage, minimal filter losses,

minimal common mode currents, and better voltage distribution in the machine stator winding.

The major disadvantage of using a passive filter in general is the higher power loss in the system. When

a filter is installed in parallel with the load, the filter current flows through the filter resistors and dissipates

power. Series reactors also might be installed on the lines to decrease the overvoltages. Although these

reactors are typically low-resistant coils, they carry the load current. They can have a high copper loss in

the windings and a high core loss due to the high-frequency pulses. They also cause a voltage drop in the

line that is typically 3-5%. Other disadvantages of installing a filter for overvoltage mitigation is the added

cost, and the need for a bigger space. Additionally, each filter has its own advantages and disadvantages

that are discussed below [86], [93–104].

 Chapter 4: Overvoltage Mitigation and Protection in VFDs 99

a. Filters on Motor Terminals

a.1. RC: low voltage overshoot, high dv/dt, high power loss

An RC filter on motor terminal matches the surge impedance of the cable and damps the voltage

overshoot. In its simplest form, it consists of Y-connected resistors and capacitors in series. Figure 4-3

shows an RC filter installed in a VFD system. RC filter on the motor terminal works like a snubber circuit.

For high-frequency components of the voltage pulse, capacitor acts as a short circuit, drawing current

through it. The energy dissipates in the resistor of the RC filter, hence damping the oscillations.

Although the RC filter looks simple and effective, there are difficulties in its design and application.

The performance of the filter depends on the cable type and characteristics. In addition, in some

applications, it is difficult, if not impossible, to access the motor terminals [100], due to the location of the

installation or other concerns. Furthermore, depending on the power loss of the filter and the installation

conditions, the cooling of the motor can be compromised if the filter installed close to the motor and in the

same location.

Figure 4-3. RC filter on motor terminal [86], [94]

Theoretically, if the impedance at the end of the cable is equal to the surge impedance of the cable, the

voltage reflection would be zero. Filter parameters are selected so that the filter impedance matches the

surge impedance of the cable at the frequency of the oscillations. Parameters of the first-order RC filter can

be calculated as below.

𝟏 𝑳
𝒁𝒇 = 𝒁𝒄 → √𝑹𝒇 𝟐 + 𝟐 𝟐
=√ Equation 4-1
𝒋𝝎 𝒇 𝑪𝒇 𝑪
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 100

Where Rf and Cf are the filter resistance and capacitance, respectively. L and C are inductance and

capacitance of the cable per unit length. One way to determine the filter parameters is to choose Rf

resistance to provide over-damping in the circuit [103]. So, we can write:

𝑳
𝑹𝒇 > 𝟐√𝑪 Equation 4-2
𝒇

L is the lumped inductance of the cable per unit.

a.2. RLC: low voltage overshoot, high power loss, low dv/dt

Similar to above, as depicted in Figure 4-4, RLC filter at motor terminal can be designed to match the

surge impedance of the cable. Adding inductors to the filter can make its response more customizable but

adds to the cost and the space needed for the installation. Different approaches of calculating filter

parameters are proposed to achieve low overshoot and minimize the power loss.

Figure 4-4. RLC filter on motor terminal [86], [94], [95], [103]

A simplified method of choosing Rf would be matching it to cable’s characteristic impedance when the

frequency tends to infinity [95]. Depending on the model used for the cable, the parameters may be a little

different.

𝒋𝝎𝒇 𝑳𝒇 𝑹𝒇 𝟏
𝒁𝒇 = + Equation 4-3
𝑹𝒇 +𝒋𝝎𝒇 𝑳𝒇 𝒋𝝎𝒇 𝑪𝒇

𝑳
𝑹𝒇 = 𝒁𝒄 = √ Equation 4-4
𝑪

This simplified method might not provide the best results, in terms of mitigating the overshoot or

minimizing the power loss. Another approach [103] chooses Rf so that it results in an overdamped circuit.
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 101

𝟏 𝑳𝒇
𝑹𝒇 < √ Equation 4-5
𝟐 𝑪 𝒇

Usually filter capacitance and inductance are chosen using computer programs, aiming at the optimal

point where the voltage overshoot and the filter power loss are in an acceptable range. Some other

constraints may be introduced based on the experience or the specifications of the components, to narrow

down the search space. For example, [103] suggest to select the resonant frequency of the filter to be five

times the switching frequency of the PWM inverter. Moreover, the operating frequency of the RLC filter

should be chosen to be near the resonant frequency to minimize the voltage overshoot.

b. Filters on Inverter Output

b.1. Reactor: high voltage drop, low dv/dt, low cost, simple

An easy and common way to increase the rise time of the voltage pulse and consequently reduce the

voltage overshoot on the motor terminals is to add a series inductor to the inverter output, as it is shown in

Figure 4-5. Inductors are usually calculated as a percentage of base impedance of the system. Inductors of 3

to 5 percent are common in industry. Although the inductors are built to have a low resistance, the load

current flows through them and the power loss might not be negligible, especially if they are installed in the

inverter cabinet. Also, inductors are bulky and need a space to be installed. Moreover, higher values of

series inductance can deteriorate the system’s power factor and cause a noticeable voltage drop [94].

Figure 4-5. Inductor on inverter output [94]

Another variant of the this filter can be used by adding a parallel resistor to the inductor to provide a

more customizable filter [101]. The design process of such a filter is simpler compared to an RLC filter,

while providing an acceptable overshoot and power loss in some applications.

 Chapter 4: Overvoltage Mitigation and Protection in VFDs 102

b.2. LC: low loss, low overshoot, complicated design due to resonance

A major disadvantage of using a passive filter is the active power loss in the filter resistor. One

approach to decrease the power loss in the filter is to remove the resistor component of the filter and only

use capacitors and inductors. An LC filter on inverter output can decrease the overvoltage effectively and

with low power loss. This filter, similar to other filters with parallel capacitors, has the advantage of

compensation of the reactive power needed by the cable and the motor. However, the design should

consider the filter resonance and its complications. These filters are often designed with a resonant

frequency, 𝑓𝑟𝑒𝑠 , significantly below the switching frequency, and above the output fundamental frequency

to avoid resonance with the load.

𝒇𝒓𝒆𝒔 = 𝟏⁄ Equation 4-6

𝟐𝝅√𝑳𝒇 𝑪𝒇

Figure 4-6. LC filter on inverter output [94], [96], [104]

LC filter commonly is installed with inductors in series and capacitors in parallel [94], [96], [104], as it

is shown in Figure 4-14. Various topologies of this filter can be used depending on whether the common

point of the capacitors are connected to the dc link or not [96]. Connection of the components determines

the maximum voltage and current of the components and also affects the performance of the filter.

Moreover, connecting the capacitors and inductors in parallel and installing them in series with the line is

also suggested [97].

b.3. RLC: high power loss, low overshoot

To avoid resonance in LC circuit, a resistor is added to the filter. Elements of an RLC filter can be

connected in different ways. Traditionally, RLC filter comprises of an L in series with the load and an RC

branch connected to the ground [86, 89], as seen in Figure 4-7(a). However, other combinations of the RLC
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 103

filter are also proposed. The disadvantageous of having the resistors in series with capacitors in traditional

topology is that they conduct the filter current, thus larger high-power resistors are required. Increasing the

resistors to decrease the current, will reduce the filter performance. To resolve this problem, resistors can

be placed in parallel with capacitor (not shown here), or in parallel with inductors, as shown in Figure

4-7(b). In an RLC filter, it is possible to connect the neutral point of the wye connection of filter to the mid-

point of the DC link to cancel the common voltage. However, in this case, circulating current that flows

through filter elements and IGBTs should be considered [100].

(a)

(b)

Figure 4-7. RLC filter on inverter output [86], [94], [95]

Selection of the resistance of the filter is a compromise between the power loss and the damping. It

should be selected to provide an effective damping while keeping the filter power loss in an acceptable

range. One approach is to select the resistance to be equal to the cable characteristic impedance at high

frequencies [95]. The reactance and the capacitance of the filter are usually calculated with the computer

programs aiming to make a balance between the power loss of the filter and the voltage overshoot at the

motor terminal.
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 104

b.4. LC clamp: medium loss, extra components

As discussed above, resistors in an RLC filter damp oscillations to avoid high overvoltages. However,

they add power loss and increase the size of the filter considerably. One alternative approach is to not use

resistors and instead clamp the voltage at the DC bus with fast recovery diodes. One topology used for this

filter is shown in Figure 4-8 [98]. In this filter, the LC resonating voltage is clamped to the dc-bus voltage,

and the rise time of inverter output voltage can be controlled by the values of L f and Cf of the filter.

Connecting the wye connection point of the capacitors to the midpoint of the DC link, also, helps to

achieve the reduction in differential mode voltage and common mode current [102]. Although this filter

would have more components compared to LC and RLC filters, its power loss is less than that of damping

resistors [86], [94].

Figure 4-8. LC filter with diode clamp on inverter output [86], [94], [98], [102]

c. Filters on Both Motor Terminals and Inverter Output

c.1. RL-plus-C: low loss, low dv/dt

An RL-plus-C filter comprises of a parallel inductor-resistor that is connected in series with the line at

the inverter output, as well as having a capacitor in parallel with the load. The basic idea of this filter is that

RL impedance is small at low frequencies and is equal to characteristic impedance at high frequencies. In

high frequencies L impedance will be high and neglected. So, we can choose Rf = ZC [93]. Filter’s

capacitor and inductor is calculated based on circuit impedances and the desired overvoltage on the load.

An RL-plus-C filter is seen in Figure 4-9.

 Chapter 4: Overvoltage Mitigation and Protection in VFDs 105

Figure 4-9. RL-plus-C filter [93]

c.2. RLC-plus-C: low loss, low dv/dt, low bearing currents

This filter can be considered a variation of RL-plus-C filter, with added capacitor to further decrease

common mode and bearing currents. It also shows better performance in reducing the overvoltage spikes

[99]. RLC-plus-C filter also can be seen as a conventional RLC filter on inverter output plus extra Y

connected capacitors on motor terminals, as it is seen in Figure 4-10.

Figure 4-10. RLC-plus-C filter [99]

4.2.2. Active Filters

Active filters are filters with both passive components and active components such as power electronic

switches and amplifiers. They are common in power systems in power quality improvement and harmonic

mitigation applications. The main concept in an active filter is to mitigate harmonics by injecting active

power with the same frequency but with reverse phase. Because of the complexity and extra components,

active filters can be expensive and can lower the reliability of the system. At the same time, they are more

flexible and generally have lower power loss compared to passive filters. Usually for low power motor
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 106

drive systems, a passive filter is a common and low-cost solution. However, for high-power systems,

investing in an active filter might be a solution to consider.

Some studies proposed to use active filters in motor drive systems to suppress the overvoltages [105]–

[108]. Various topologies of the active filters are proposed. Generally, active filters are comprised of

parallel passive elements, a common-mode transformer to add a positive or negative voltage, a dc source

and electronics switches to control the filter behavior. A general active filter used on the inverter output is

shown in Figure 4-11. An active filter on inverter output

Figure 4-11. An active filter on inverter output [107]

This filter is effective in reducing the common mode voltage, common mode current and EMI. Z 1 and

Z2 serve as common mode voltage detection circuit, where it used to control the switches and inject

compensating voltages to the lines by the common mode transformer.

Some filters may include complex sampling, processing and control units. While more switches and

more processing generally provide better results and lower power loss, these solutions may cost more and

have lower reliability overall, due to extra parts adding to the possibility of the failure. An example of such

a filter is shown Figure 4-12. This filter uses an energy-recycling module (ERM) to reduce the power loss

in the previously proposed filter (also shown in the same figure), while effectively reducing the

overvoltages on the motor terminal. Figure 4-13 shows the line to line voltage at the inverter output and the

motor terminal. It is seen that peak value of the overvoltages is reduced from 570 V to 325 V, where the
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 107

pulse amplitude is 300 V at the inverter output. Power loss was 18% of the load power with a resistor used

in the filter, instead of ERM. Although this is a huge power loss in the filter and requires a high-power

resistor or possibly using a costly cooling system, still it is less than the power loss in the conventional RC

and RLC filters. Using the proposed ERM, [105] reduces the power loss of the filter down to 2% of the

load. Nevertheless, this reduction in the power loss comes at the expense of using an extra 6 power

switches, 3 inductors and one common mode transformer, as well as higher complexity of the control

system.

Figure 4-12. Active filter with energy-recycling module (ERM) proposed in [105]
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 108

(a) (b)
Figure 4-13. Reduction of the overvoltages in a drive system with an active filter proposed in [105]:
(from top) voltage at inverter output, middle of the cable and motor terminal (a) without the filter (b)
with the filter

These components are an addition to the original active filter that used 10 capacitors, 3 inductors, 3

resistors, 6 power switches and 6 rectifier diodes. In practice, these extra components add to the cost of the

filter and reduce its practicality in applying to a wide range of applications.

4.2.3. PWM Control

Fast rising PWM pulses causes reflection overvoltage on motor terminals. But at the same time, PWM

technique provides immense flexibility in control of the motor. This capability of precise PWM control has

also been used to solve the overvoltage problem. The target is to shape the inverter output voltage so that

the rate of change in the filter output during voltage transitions becomes lower. Similar to active filter

method, this method needs passive components as a filter, and it may need a feedback circuit and sensors.

Also, complexity of the control scheme is a disadvantage. However, its advantage is achieving the goal of

active filters in reducing the motor overvoltages, using the existing power switches in the inverter.

Researchers have proposed different methods on utilizing the controllability of the inverter itself in

overvoltage mitigation [106], [109]–[111]. Reference [106] uses an active dv/dt control to eliminate the

overvoltages completely. Figure 4-14 and Figure 4-15 show the topology and the fundamental idea of their

method. In this approach, a dead time is applied after the rising edge and before the falling edge of the

pulse. After applying the pulse for the half of the rise time, it is reverted for the second half, reducing the

average amplitude of the pulse to half. Thus, it will create a voltage reflection with the amplitude of one per

unit. This is the same as supplying 2 pulses with a voltage of half the amplitude and duty cycle of 50%.
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 109

Output voltage of the filter will reach 1 pu at t = τ. At this moment no transient occurs, since the step

voltage and the filter output voltage are the same. This approach needs an LC filter on inverter output.

However, filter is designed for higher cutoff frequency and makes it possible to use smaller filter

components.

Figure 4-14. LC filter used in [106] to achieve active filtering with PWM control

Inverter output
Filter output

With active control

Without active control

Figure 4-15. Active dv/dt control for overvoltage mitigation, (top) inverter output, (bottom) motor
terminal [79]

The above mentioned method assumes the load current to be zero. However, in real conditions, motor

current will affect the charging and discharging sequence of filter capacitor. Inductor current will increase

during the first half of the rising time when switches are on. Then, for the second half, switches turn off and

filter current circulates through switch body diode. When there is a load current, the filter current needs to

change direction in the charging and discharging sequence. In order to restore the current of the filter
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 110

inductor to the load current level, the power switch parallel to the diode that was conducting must be turned

on at the point when the diode stops conducting. This adds to the complexity of gate pulse control.

This situation is depicted in Figure 4-16. Assume the load current is negative and the positive pulse

edge occurs. When S1 turns on, the filter inductor’s current on phase A increases. For the second half of the

rise time, S1 switch should turn off and the filter current should decrease and reach the previous stable

point. Current flows through the body diode of the S2 switch until it reaches zero. At this moment, switch

S2 should turn on to provide a path for the circulating current. This is what is seen in the figure below for

the lower switch. Turning on the switch in the same leg increases complexity of the control, since cautious

measures should be considered for switches, not to be ON at the same time.

Figure 4-16. PWM control pulses when the load current is negative

As it is mentioned in Chapter 3, in a VFD system with a long cable and high switching frequencies it is

possible that the amplitude of overvoltages exceeds 2 pu and even reach 4 pu, where 1 pu is the DC link

voltage. Various reasons can create overvoltages greater than the dc bud voltage, including double pulsing

and double transition. Double pulsing is when a pulse is applied shortly after another pulse and before the

oscillations from the first pulse has been damped. Superposition of the oscillations from 2 pulses can create
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 111

overvoltages higher than 2 pu. The oscillations from the rising and falling edges of one pulse can also

create a similar phenomenon. Double transition or polarity reversal occurs when the voltage of a line

changes from –Vdc to Vdc. This is a pulse with an amplitude of 2Vdc that can create overvoltages of up to 3

Vdc (i.e. 4 Vdc from -Vdc).

A simple way to eliminate the over-2pu overvoltages due to double pulsing or short pulses is

suppressing any pulse with width under a certain limit, tmin, to allow the oscillations to fully decay and

eliminate the possibility of superposition of voltage reflection waves [7]. Another innovative way for

overvoltage mitigation uses the oscillations from 2 near-by pulses, where the superposition is possible, and

adjust the pulse width so that the oscillations from the next pulse event or the falling edge of a pulse

superimposes the ones from the previous pulses or the rising edge of a pulse with a negative amplitude,

offsetting a certain amount of the overall overvoltage oscillation [110]. The principle of this method is

shown in the Figure 4-17.

(a) (b)
Figure 4-17. Overvoltage oscillations from (a) switching on (b) switching off

Assume the rising edge of a pulse is on P1. If the falling edge P4 occurs at P3, the overvoltage

oscillations from both events will boost each other and the overall overvoltages may go beyond the 2 pu

limit. However, a Double Pulsing Offset Technique (DPOT) PWM control adjusts the adjacent pulsing

events so that the falling edge P4 occurs on either P6 or P7. This way oscillations from one event will be

offset by the other.

Another PWM control technique recommends avoiding double transition or polarity reversal (pulses

from –Vdc to Vdc) [7]. Selecting the switching frequency with regard to the system frequency response is a

crucial step in the VFD design. Considering the components, the switching frequency should be selected to
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 112

be far from the resonance frequency of the system [109]. These cautions along with the previously

mentioned ones will limit the overvoltages to at most 2 Vdc.

4.2.4. DC Cable between Rectifier and Inverter

As it established already, a typical VFD creates multiple problems in the system, most of which are

associated with the reflection phenomenon in AC long cables. It is possible to eliminate the voltage

reflection overvoltages in AC drives using a DC long cable between the rectifier and the inverter, instead of

using an AC long cable between the inverter and the motor. This approach is illustrated in Figure 4-18.

The main advantage of using a dc cable is obvious: no reflected wave. That also translates to no

overvoltages, smaller common mode currents, smaller cable charge current, and no filter needed. There are

other advantages, such as using two long cables instead of three, and low voltage drop on the cable. In very

long cables, such as submarine cables, the reactive power of the cable itself, occupies the cable current

capacity. Using a dc cable instead can eliminate the reactive power of the cable. However, using a dc cable

has its own disadvantages, too.

Short Circuit protection is a fundamental aspect of every electrical system. The proper functioning of

most of the traditional relays and fuses is based on having a zero crossing in the current. Since the dc

current does not have a zero crossing, short circuit fault protection in DC link are more difficult and need

more expensive protection equipment. Other disadvantages of installing a long dc cable include separate

installation of the inverter and the rectifier that might not be possible in all situations, costly maintenance,

and higher harmonic currents and losses [68].

Figure 4-18. Long DC cable connection between the rectifier and the inverter [112]

Other than technical advantages and disadvantages, the widespread use of a system depends on the cost

analysis of that system. In regard to the above mentioned DC-Cable VFD system, it is shown that a
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 113

noticeable saving can be achieved in terms of copper volume and cost [112]. At the same time, installing a

DC capacitor bank across the inverter terminals is necessary to eliminate the DC transient overvoltage

peaks at the inverter. Extra cost of the DC circuit breakers should also be considered.

4.2.5. Energy Varistor

Conventionally, varistors are used to protect against surges and overvoltages that occur limited times

during their lifetime. Generally, during a discharge, an MOV conducts a large current and heats up for a

short time. But since the event is not persistent, the MOV has time to cool down to the environment

temperature and be ready for the next event. When an MOV is used to mitigate the repeating high

frequency overvoltages, the MOV repeatedly discharges high-peak impulse currents when each pulse is

applied. Also, after each pulse, it has a short time to cool down until the next pulse (depending on the

switching frequency). Manufacturers do not design MOVs for repetitive pulse discharge. There are no

application notes or standards on how to select the MOV for these situations. The effects of the repeating

pulses on the MOV’s performance and lifetime are unknown. However, the possibility of the concept is

proven by a research experiment that uses energy varistors in a long-cable motor drive system on motor

terminals [113]–[116]. These experiments show the effectiveness of MOV application for overvoltage

mitigation and serve as a proof of concept. Also, there are attempts to show robustness of the MOVs by an

accelerated test of 1000 hours in 140 °C, 55 degrees above the maximum operating temperature of the

MOV. However, this experiment provides extra cooling using an aluminum plate attached to the cupper

electrodes.

Figure 4-19(a) shows the reduction of the overvoltage with application of MOVs on motor terminals in

line to ground connection. The DC link voltage is 600 V and the rise time of the pulse at inverter output is

100 ns. Cable length is 130 m. Without a varistor, the peak voltage on motor terminal reaches to 1200 volts.

Installing MOVs on motor terminal, reduces this voltage to 1020 volts. The power loss in the MOV

depends on the amplitude of overvoltages and its ratio to the MCOV of the varistor as well as switching

frequency. Figure 4-19(b) shows the power loss for an energy varistor for different switching frequency
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 114

and different cable types. Also, MOV cooling effect is considered in this figure. For extra cooling, an

aluminum plate is attached to copper electrodes to provide better heat dissipation. It is seen that for a given

condition, an MOV can be used for overvoltage mitigation up to a certain switching frequency. Cable

shield should also be taken into account. For example, this set up can work continuously up to switching

frequency of 8 kHz, without any extra cooling. See pink points in Figure 4-19(b).

(a) (b)
Figure 4-19. Overvoltage mitigation with MOV [113]

Installing MOVs on motor terminals affects overvoltages in two distinct ways. One effect comes from

the capacitance of the varistor. Varistors’ capacitance at the end of the cable act like a filter and reduces the

overvoltages. If the varistors MCOV is close to peak value of the overvoltages, the resistive current will be

negligible and the capacitance of the varistor is the only limiting factor of the overvoltages. In this

condition, a capacitive current will flow through the MOV. The rise time of the voltage at motor terminals

also increases due to the change in capacitance of the circuit. Another effect that limits the overvoltages at

motor terminal is due to varistor’s voltage clamping aspect. When an overvoltage amplitude increases, a

resistive current also flow through the varistor in addition to the capacitor current discussed above. This

limits the voltage on motor terminal which is connected parallel to the MOVs. The higher the amplitude of

the overvoltage compared to varistor’s MCOV, the higher the resistive current. It is the resistive current

that causes the power loss in the varistor and heats it up. Figure 4-20 shows overvoltage mitigation by two

varistors with different MCOV for a motor drive system with the DC link voltage of 800 V. The amplitude

of voltage peak at motor terminal is 1600 V. In Figure 4-20(a) varistor is chosen so that it conducts a very
 Chapter 4: Overvoltage Mitigation and Protection in VFDs 115

(b)
(a)

Figure 4-20. Current and voltage of the MOV [114]

small resistive current (MCOV is close to voltage peak). Voltage peak is decreased to 1400 V. Here, the

mitigation is only caused by varistor’s capacitance. In Figure 4-20(b) the varistor draws about 12 A current

at the peak of the voltage. This is the peak value of the resistive component of the varistor current. In this

condition, overvoltage peak value drops to 1100 V. In this condition, the power loss in the varistor depends

on the amplitude of the current and voltage as well as the switching frequency.
 Chapter 5: MOVs in Variable Frequency Drives 116

Chapter 5:

5. MOVs in Variable Frequency Drive

Systems

In this chapter, MOV behavior in VFD systems is investigated. Conventionally, MOVs are designed to

discharge infrequent currents due to lightning and occasional power system switching surges. When MOVs

are used in VFD systems, they may be exposed to repetitive pulses created by switching. These pulses can

discharge repetitive currents through MOVs and cause over-heating and eventually failure of the MOV. To

prevent this, designers have to select MOVs with higher voltages to ensure that MOVs can tolerate the

overvoltages. This approach diminishes MOVs capability and performance to protect the system

efficiently, since appliances are subjected to higher overvoltages. There has been limited research and

publications on the topic to explain the phenomena. This chapter provides test results and discussion on the

behavior of the MOVs when repetitive pulses of different amplitude and duration are applied. Furthermore,

new approaches are proposed to allow MOVs to be safely operated in the VFD systems, while providing a

proper protection level for the motor and the VFD system.
 Chapter 5: MOVs in Variable Frequency Drives 117

5.1. MOV installation on VFD systems

In general, MOVs provide a low impedance path to the ground for the surge current. Typically, MOVs

are installed between lines and ground. When there is access to the neutral wire, additional MOVs might be

placed in line to neutral mode. Occasionally, extra MOVs might be placed in line to line mode to provide

an even better protection. Considering the cost and the sensitivity of the equipment and the surge

environment where the system is located, the surge protection designers might choose to use a 3-mode, a 4-

mode, a 7-mode, or a 10-mode protection scheme [117]. These combinations are shown in Figure 5-1.

Depending on where the MOV is installed on the VFD cable, it may be subjected to different voltage

amplitudes and waveforms. Section 3.4 describes the voltages in the VFD systems in details. The most

common location to install the MOV in the VFD system is on the motor terminal, between line and ground.

Some SPDs provide extra protection by adding line to line and line to neutral MOVs.

Figure 5-1. Modes of surge protection in 3-phase systems

5.2. Behavior of MOV under Repetitive Pulses

Traditionally, MOV manufacturers test some samples for standard surges for a limited number of

impulses. Typically, MOVs are not designed, nor tested for repetitive high frequency pulses. With

expansion of sensitive electronic devices to many fields and applications, MOVs are going to be used in

systems with non-traditional voltage waveforms. One of these new applications is using MOV-based SPDs

to protect VFD controlled motors. In these systems, MOVs are needed to be installed to protect the cable

between the inverter and the motor against lightning surges. These MOVs are subjected to a PWM voltage.
 Chapter 5: MOVs in Variable Frequency Drives 118

Little studies have been conducted on the behavior of the MOVs under such voltages. As a result, engineers

usually over-design and install MOVs with a higher MCOV to avoid any possible failure. This practice

reduces the protection performance and puts the expensive and sensitive appliances at risk.

As previously stated, in a VFD system, voltage pulses at inverter output are clean with minimum

distortion. On the other hand, pulses at the motor terminal are distorted with high spikes. This is shown in

Figure 5-2. The amplitude and the frequency of the spikes depend on the length of the cable and the

specifications of the inverter. For a motor drive system with a long cable (l cable > 10m) surge protection on

both ends of the cable is recommended by IEEE [91]. In such a configuration, although both SPDs are

installed on the same line (one on the inverter output and the other on motor terminals), the voltage pulses,

conducted currents and the power loss that they experience are considerably different.

In order to investigate the behavior of the MOVs under voltage pulses, this chapter experimentally

injects pulses from a power supply, as well as from a VFD inverter. Pulses from the power supply have

long rising and falling times (100 ~ 150 µs). Pulses from the VFD have fast rising and falling edges (100 ~

150 ns).

Figure 5-2. PWM pulses in a VFD system

5.2.1. MOV’s Response to slow Pulses from a power supply

In this experiment, MOVs are subjected to millisecond range pulses from a voltage source to study their

behavior under pulses. A programmable Chroma Power Supply 61504 is used to create pulses with pre-

determined width, amplitude and number, in this experiment. Figure 5-3 shows the test configuration. The
 Chapter 5: MOVs in Variable Frequency Drives 119

voltage source has maximum pulse voltage of 424 V, with 2 kVA of apparent power that can deliver about

50 A for a short time in ms order. The voltage rise is slow, and the rise time of the pulses is about 100 µs.

The waveform is shown in Figure 5-4. The pulse waveform is not exactly square and has an overshoot of

about 10 V. The rise time, the time required for a pulse to rise from 10 percent to 90 percent of its steady

value, is 110 µs.

Figure 5-3. Pulse application to 150 TPMOV with the programmable Chroma power supply

350
300
250
Voltage (V)

200
150
100
50
0
-0.5 -50 0 0.5 1 1.5 2 2.5
Time (ms)

Figure 5-4. Pulse waveform applied with the programmable Chroma power supply for a 300 V pulse

Measurement of the MOV current is a difficult task. The normal leakage current in voltages below its

MCOV is less than a milliampere, with its resistive component being around 100 µA. For higher voltages

MOV current can increase rapidly and reach several amperes for a short time.

For microampere range, having a 100-ohm current sensing resistor in series with the MOV is acceptable

and can provide a good signal to noise ratio to avoid noise and errors in measurements. However, for

higher currents in the range of several ampere, the 100-ohm resistor will cause a large voltage drop in the
 Chapter 5: MOVs in Variable Frequency Drives 120

applied voltage to the MOV. One way to alleviate this challenge is to change the sensing resistors based on

the maximum current. Another approach is to use clamp current probe with an electronic amplifier. In this

experiment two Tektronix TCP312A current probes are used. One has 30 A max current with a frequency

range of dc to 100 MHz. The other one is Tektronix TCP0150 with 150 A max current and frequency range

of dc to 20 MHz that are used to measure currents higher than 30 A.

Since the MOV has a parasitic capacitance, every change in its voltage will cause a current through it.

We call this part of the MOV current, the capacitive component. Two spikes of capacitive currents are

shown in Figure 5-5. Zooming into the rising edge of the pulse shows that a peak of 15.2 mA current as a

result of 50 V change in the voltage in a time span of 25 µs. This gives an approximate capacitance of 7.6

nF.

𝒅𝒗 𝑰𝑪
𝒊𝒄 = 𝑪 → 𝑪 = ∆𝒕 → 𝑪 = 𝟐𝟓 𝝁𝒔 × 𝟏𝟓. 𝟐 𝒎𝑨 ÷ 𝟓𝟎 𝑽 = 𝟕. 𝟔 𝒏𝑭 Equation 5-1
𝒅𝒕 ∆𝑽

This value is consistent with our measurements for the MOV with a precision LCR meter, as it is shown

later in this chapter.

VMOV

IMOV

Figure 5-5. 150 V Pulse applied by the power supply to a 150TPMOV and MOV current

Assume a simple model of a constant capacitor in parallel with a non-linear resistor for the MOV. The

current between two edges of the pulse, where the voltage is almost constant, here called the resistive
 Chapter 5: MOVs in Variable Frequency Drives 121

component, is responsible for power consumption and heating. It is small when the voltage pulse amplitude

is in proximity of MCOV, thus it is not accurately measurable with our current probe. Increasing the pulse

voltage, increases both capacitive and resistive components of the MOV current. Capacitive current

increase is proportional to voltage increase, as ΔV/Δt increases. It is expected considering Equation 5-1.

Resistive current, on the other hand, increases exponentially with increasing voltage and its behavior is

affected by multiple factors. For example, Figure 5-6 shows the MOV current for 260 V pulse. The

capacitive peak current has increased to 28 mA, as expected from voltage change. Resistive current, is

negligible for voltages up to about 200 V and then increases nonlinearly with voltage increase. The peak of

current corresponds to the peak of voltage. Then, current decreases while the voltage amplitude is almost

constant.

Figure 5-6. 260 V Pulse applied by the power supply to a 150TPMOV and MOV current

(a) (b)

Figure 5-7. (a) MOVs used in the experiment with and without case (b) MOV's structure
 Chapter 5: MOVs in Variable Frequency Drives 122

There are multiple factors affecting the waveform of the MOV’s current. First, the inductance of the

ceramic itself. Although it is small, it causes a small delay in rising the current. The waveform of the

voltage pulse also plays a role in the waveform of the current. The pulse has a 10 volts overshoot that

corresponds the peak value of the current, then both voltage and current decrease. However, after the pulse

voltage settles and remains constant, the current continues to decrease. This should be justified by taking

the MOV’s microstructure into account. This is similar to the behavior of the MOV when a DC voltage is

applied. The leakage current starts from a higher value and decreases until it settles on a value. That is why

for the MOV’s leakage current measurement, it is instructed to read the value 2 seconds after applying the

voltage. Another factor that also plays a role in the current waveform is the heating of the MOV. The

current of an MOV increases with increasing temperature for a given voltage. For the applied pulses, in

lower currents, where MOV does not heat up, the current decreases after reaching its peak. For higher

currents, the heating effect has higher impact and we can see the current increases. This can be seen in the

below figures. Figure 5-8 shows the waveform of the MOV current for four pulses with different

amplitudes. As it is seen in Figure 5-8(a), the peak value of the current is 2.5 A. The MOV current

decreases after reaching its peak. On the other hand, in Figure 5-8(d) the current peak is 34 A. Here heating

of the MOV during the pulse increases its current.

(a) (b)

(c) (d)

Figure 5-8. Current waveforms of MOV for 1 ms voltage pulse (a) 2.5 A (b) 17A (c) 27 A (d) 34 A
 Chapter 5: MOVs in Variable Frequency Drives 123

For pulses in-between the two, the effect of current decrease due to microstructure and the effect of

current increase due to the heating affect the waveform. Depending on what factor has more influential

effect the current might increase, decrease or remain constant, as it is seen the Figure 5-8(b) and Figure

5-8(c).

To investigate the behavior of the MOVs under millisecond-range pulses further, 3 MOVs are selected

to be subjected to pulses with various width. These MOVs have the same MCOV of 150V, but different

nominal voltages, all of which are within the standard range. They are listed in the Table 5-1. The number

of pulses applied, their width and the interval between them are also listed in Table 5-2. The amplitudes of

the applied pulses, start from 150 V and increases in steps of 10 V, until the power supply reaches its

current limit of 50 A. The distorted pulses, where the distortion is clearly from power supply current limit,

are discarded from the analysis. In total, 200 pulses are applied to 3 MOVs, and the results of each

experiment are compared to others to find out if pulse width and interval between pulses have any effects

on the waveform and peak value of the MOV current. The capacitive current is neglected and only the

resistive current is considered.

Table 5-1. MOVs used in pulse experiment and their nominal voltages

MOV code V1mA (V)

MOV230 P150-13 230
MOV240 P150_12 240
MOV250 P150_15 250

Table 5-2. Number of pulses, their width and intervals applied to MOVs
Experiment Number of Pulses Pulse Amplitude Pulse Width Interval
code in the sequence (V) (ms) (ms)
CP1 1 150 to 370 0.4 -
CP2 1 150 to 350 0.7 -
CP3 1 150 to 350 1 -
CP4 1 150 to 340 1.5 -
CP5 2 150 to 370 0.4 3.6
CP6 2 150 to 370 0.4 0.2
CP7 2 150 to 350 1 0.2
CP8 10 150 to 300 4 36
 Chapter 5: MOVs in Variable Frequency Drives 124

The first expected result of the experiment is that V1mA of the MOV is the major factor in the MOV

current. For example, as it is seen in Figure 5-9, for a 310 V pulse, MOV230’s peak current is 32.8 A. This

value for MOV240 and MOV250 is 26.8 and 6.0 amperes, respectively. Furthermore, we should notice that

the current delivering capability of the power supply is limited, and it has an internal impedance, too. When

applying the pulse by a supply, MOV will limit the output voltage by discharging a larger current. Here,

although the settings of the pulse are the same for these three, the pulse voltage is not. For MOV230, where

the supply delivers a large current, the pulse voltage has dropped to 294 V. For the MOV240, the pulse

amplitude is 302 V, and for MOV250 it is 314 V. If the supply had higher short circuit current capability,

MOV 230 would draw even larger current from the power supply. Applying multiple pulses with different

amplitude, we can measure the peak current for different voltages and draw the curves in Figure 5-10.

MOV230 MOV240 MOV250

Figure 5-9. MOV current for a 310 V pulse

MOV230 MOV240 MOV250 MOV230 MOV240 MOV250

MOV current Peak (V)

50 50
MOV current Peak (V)

40 40
30 30
20 20
10 10
0 0
150 200 250 300 350 400 150 200 250 300 350 400
Pulse voltage setting (V) MOV Voltage Peak (V)

(a) (b)

Figure 5-10. MOV peak current vs. (a) pulse voltage setting (b) MOV voltage peak
 Chapter 5: MOVs in Variable Frequency Drives 125

Figure 5-10(a) shows the current peak vs. the pulse voltage setting (pulse amplitude if the output of the

power supply is open). This chart shows what MOV current to expect, if we know the open circuit voltage.

For example, if we know the DC link voltage in an inverter, this value will also be equal to open-circuit

pulse voltage amplitude. However, this chart is dependent on the impedance of the power supply. A more

useful chart would be the one shown in Figure 5-10(b) that shows the MOV current vs. MOV voltage.

These charts approximately match the datasheet VI characteristics. However, there is an important

distinction: Typically, when referring to the MOV’s VI characteristics, the upper voltage limit is important

for currents higher than 1 mA. That is why the manufacturers do not specify the lower limits for Amp-

range currents, as they do for sub milliampere currents. This is because traditionally engineers only care

about MOV’s leakage current. For higher currents, including tens of amps, the only traditionally important

factor is the upper limit of clamping voltage. However, for application of the MOVs in VFD systems, they

may conduct currents in the range of several amperes. It becomes important to know the highest current

that an MOV will draw for a specific voltage. This is the lower limit curve that is added to Figure 5-11, just

to show the concept.

Figure 5-11. VI characteristics of MOVs used in the experiment [58]

 Chapter 5: MOVs in Variable Frequency Drives 126

Further experiments are conducted to study the relation of pulse width (in milli-second range) and the

MOV current. For this purpose, the above experiment is repeated for different pulse widths, as seen in

Table 5-2. The shortest pulse width that the power supply can create is about 0.4 µs. Thus, it is decided to

apply pulses with width of 0.4, 0.7, 1.0 and 1.5 µs. Waveforms for the 300 V pulse applied to MOV230 are

seen in Figure 5-12. It is clear that pulse width affects the maximum peak value of the current. If the pulse

duration was enough for the MOV current to reach its peak, and also if the MOV current amplitude was

constant after reaching its peak, MOV current would be independent of the pulse width. Since it takes some

time for the MOV’s resistive current to rise, so the current peak is smaller for the short pulses. However,

the second factor can affect the current both ways. It is described before that the MOV current during the

pulse can be increasing, constant, or decreasing, depending on the amplitude of the MOV and its heating

specifications (see Figure 5-9). Thus, if the current is increasing the peak value of the current will be higher

for wider pulses, as it is seen in Figure 5-12. On the other hand, if the current is decreasing during the

pulse, as it is seen in Figure 5-13, it is not the case.

1.5 ms
1.0 ms
0.4 ms

0.7 ms

Figure 5-12. Pulse voltages and MOV currents for 300 V pulses for MOV230
 Chapter 5: MOVs in Variable Frequency Drives 127

Figure 5-13. Pulse voltages and MOV currents for 310 V pulses for MOV250

It is seen in Figure 5-13 that 0.7 ms pulse is wide enough to allow the current to reach its peak. For

pulses wider than that, the current peak is the same, although the average current and the average power is

decreasing. For pulses narrower than 0.7 ms, the MOV current does not reach its peak, thus, has a smaller

peak value.

Another question that can be asked after reviewing these results is whether adjacent pulses can affect

the MOV current. For a resistor, two identical pulses create two identical current waveforms, but this is not

true for an MOV. Because of the MOV’s microstructure, it takes some time for the current to rise. Also, for

a constant dc voltage, the current starts to fall and settles after some time. Experiments CP5 to CP7 intend

to investigate this topic. It is found that the pulses can affect the MOV’s current for the subsequent pulses.

Figure 5-14 shows an oscilloscope screen capture of two pulses of 300 V and 1 ms, applied to

MOV240. The interval between pulses is set to 0.2 ms. It is clearly seen that the MOV current during the

second pulse does not rise as fast as it does during the first pulse. This effect is more seen in pulses close

together, and less in the pulses that are apart from each other.
 Chapter 5: MOVs in Variable Frequency Drives 128

Figure 5-14. two pulses of 300V, 1 ms, applied to MOV240

(a) (b)
Figure 5-15. two 310 V, 0.4 ms pulses applied to MOV230 with interval of (a) 0.2 ms (b) 3.6 ms

Figure 5-15 shows MOV230’s current for two pulses of 310 V, 0.4 ms long. Both first pulses discharge

28.8 A currents. However, the MOV currents during the second pulses are different. When the interval

between pulses are large, the second pulse’s current is almost as large as the first pulse’s one, but when the

pulses are close to each other, the MOV’s current during the second pulse reaches 26 A which is almost 2

amperes less than the first one.

In the final experiment in this section, ten pulses with 4 ms width and 36 ms interval are applied to a

150V MOV. The amplitude of pulses is changed from 150 to 300 V. In all amplitudes, it is seen that the

MOV conduct higher current during the first pulse, and it decreases for the subsequent pulses. Pulses of

290 V are seen in Figure 5-16. More details can be seen in Figure 5-17, where the 1st, the 5th, and the last

pulse are shown.

 Chapter 5: MOVs in Variable Frequency Drives 129

Figure 5-16. ten pulses of 290 V and 4 ms applied to an MOV

Figure 5-17. MOV current during ten consecutive pulses (a) 1st (b) 5th (c) 10th

It is seen that the current peak for the first pulse is 2.32 A, where it drops to 1.55 A and 0.98 A for the

5th and 10th pulses. Not only the peak value of the MOV current decreases in consecutive pulses, but also

the steady state part of the current, here called I ss, also decreases. These parameters are depicted in Figure

5-18 for 290 V pulses.

2.5
Ipeak (A)
2 Iss (A)
Current (A)

1.5

0.5

0
1 2 3 4 5 6 7 8 9 10
Pulse number

Figure 5-18. MOV's peak and steady state currents for ten consecutive pulses
 Chapter 5: MOVs in Variable Frequency Drives 130

5.3. Behavior of MOV under PWM Pulses

When an MOV is installed to protect the cable that is connecting an inverter to a motor, it is subjected

to a PWM voltage. On the inverter output these pulses are clean with almost no overshoot. On the motor

terminal, the waveform depends on the length of the cable. It can be relatively clean PWM with small

voltage overshoots for short cables, or there can be high frequency overvoltages occurring on all edges of

pulses. In this section we are going to see how MOV behaves under PWM pulses. This sectin shows that

the capacitance of the MOV has a noticeable effect on the overvoltages. Adding an MOV to the end of line,

regardless of its nonlinear resistance and because of its capacitance, increases the period of overvoltage

oscillations and reduces their amplitude. Also, it is discussed how to calculate the resistive current from

voltage and current waveforms. Suggestions are made on how to choose a proper MOV.

Typically, the rise time of a PWM inverter with IGBT switches is in sub-microseconds. Such fast-

changing voltages can discharge large capacitive currents through the MOV. A simple calculation below

shows how large MOV’s typical capacitive current can become. Although large capacitive currents do not

cause any power loss or heating in the MOV, they flow through the cable and cause power loss and voltage

drop in the cable and take up its current capacity. Large currents also may trigger the over-current

protection relays and cause unwanted trips.

Rise time = 200 ns

Pulse amplitude = 340 V => i C = C dV/dt = 6.8 A

MOV capacitance = 4 nF

For this experiment, three Y-connected MOVs are installed on the 3-phase motor terminals. The common

terminal of the MOVs are connected to the ground. The test setup is shown in Figure 5-19. The current and

the voltage of the MOVs are measured and captured by a Tektronix oscilloscope.
 Chapter 5: MOVs in Variable Frequency Drives 131

Figure 5-19. MOVs connection for motor protection

Vag @ Inverter output

Vag @ Motor terminal

MOV current

Figure 5-20. Line to ground voltages and MOV's current

(a) (b)

Figure 5-21. (a) Motor current (b) MOV and Motor currents superimposed

Figure 5-20 shows an MOV’s current, connected between a line and the ground, on the motor terminal.

As it is seen from the figure, MOV conducts large currents, mainly capacitive, at each pulse edge. The peak

value of the current is 4.7 A for this MOV. The motor’s current is an ac current with the maximum value of
 Chapter 5: MOVs in Variable Frequency Drives 132

about 2 A. The peak value of MOV’s current is more than twice that of the motor. The load current for the

cable (MOV plus motor), reaches more than 7 A. Setting of the over-current relay on the inverter side

should consider these high peak currents. In fact, in the lab setup, these high currents tripped the drive

several times, so we had to change the over-current protection settings.

Other than drawing large capacitive currents from inverter, MOV’s capacitance has another effect on

the VFD system, too. Installing an MOV changes the waveform of the overvoltages, because of its large

capacitance. Taking the setup in the lab, the cable’s capacitance is 130 pF/m. For the 29-meter cable, the

total capacitance comes to 3.77 nF. According to its datasheet, A 150 V MOV’s capacitance is about 4.8 nF

at 1 KHz. This large capacitance at the end of the cable changes the reflection coefficient and the waveform

of the voltage oscillations. The line to ground waveform after connecting an MOV is seen in Figure 5-22.

The frequency of the oscillations has decreased, and the period has increased to 1.5 µs from 0.8 µs.

Figure 5-22. Line to ground voltage after installing an MOV

This can be verified in the simulation, too. Adding a large capacitance to the end of the line, increases

the period of the oscillations. Below in Figure 5-23, the same VFD system from section 3.5.6 is used. Three

4.8 nF capacitors are connected on the motor terminal between lines and the ground. It is seen from the

simulation result in Figure 5-24 that the oscillations period has increased to 1.506 µs. Also, the current

flowing through the MOV’s capacitance has the peak value of 5.45 A, which is 13% less than the measured

value of 6.2 A, shown in Figure 5-22.

 Chapter 5: MOVs in Variable Frequency Drives 133

Figure 5-23. VFD system simulation with MOV's capacitance

1.506 µs

Figure 5-24. Simulation results: MOV's voltage and capacitive current

Another important aspect of MOV’s behavior is its resistive current. As it is stated in the previous

section, the capacitive current is proportional to the voltage change rate, and the resistive current is

dependent on the MOV’s voltage amplitude. For the voltage below the MOV’s MCOV, the resistive part of

the current is small and negligible.

A simple way to measure the peak value of MOV’s resistive current is to take the current value, when

the voltage waveform’s derivative is zero, at its peak. Since the capacitive current is proportional to dV/dt,

its value at this moment is zero. In Figure 5-24, it is seen that for only a capacitor, the momentary current at

the peak of the voltage, i.e. the resistive current, is zero. For a real MOV, the resistive current is not zero,

but for voltages below MCOV it is small. Figure 5-25 shows the current of a 420V MOV for a pulse. The

peak value of the line to ground voltage is 390 V. As it is seen from the figure, for this voltage, the resistive

current is almost zero and the current waveform is zero at the peak of the voltage.
 Chapter 5: MOVs in Variable Frequency Drives 134

Figure 5-25. 420V MOV's voltage and current

Figure 5-26. 150V MOV's voltage and current

Voltage and current waveforms of a 150V MOV is shown in Figure 5-26. The peak value of the voltage

is 320 V. As seen in the figure, the resistive current at the peak moment of the voltage is not zero. The

resistive current’s peak at this moment is about 3 A. This resistive current causes power loss and heating of

the MOV. If the average power loss is higher than the MOV’s heat dissipation capability, its temperature

will rise. If there is a thermal protection, it will open and MOV will be disconnected, as it is in the lab

setup. If there is no thermal protection, the temperature rise can reach a thermal runaway point and will

lead to MOV’s failure. So, it is important that designers of MOV application in non-conventional

applications measure the resistive current and make sure the average power loss does not exceed the power

dissipation capability of the MOV. Since this power loss occurs in a pulse, obviously the switching

frequency of the PWM voltage (more accurately, number of pulses in a second) also should be taken into

account. As a general rule, this resistive current should preferably be negligible to have a desirable safety

margin.
 Chapter 5: MOVs in Variable Frequency Drives 135

5.4. High Frequency Model of MOV

In today’s industry, MOVs are not designed for high frequency voltages and there are no standard high

frequency tests that manufacturers should comply. In datasheets there are no information on high frequency

specs and capabilities of the MOV. However, there are a few academic publications on some aspects of the

high frequency behavior of MOVs [15], [118]–[123].

One of the early efforts to characterize the low voltage MOVs under high frequencies is reported in

[122], where it is showed the resistance of an MOV decreases almost linearly with the inverse frequency.

Adding this statement to the known fact that MOV’s impedance is dependent on the current flowing

through it, and the leakage current of the MOV is dependent on the temperature, they proposed a model for

higher frequencies, as shown in Figure 5-27.

Figure 5-27. MOV’s frequency- and temperature-dependent model

Dependence of the MOV’s impedance on temperature is measured with DC leakage current

measurement. The dependence on the frequency is measured by applying sinusoidal voltage to the MOV

with a signal generator (20 V peak to peak) and sweeping the frequency.

In recent years, other models are also proposed to provide a better accuracy for high frequency. Four of

the models are seen in Figure 5-28. The primary elements of these models are the same, a non-linear

resistor in parallel with a capacitor. The other elements are added to account for the parameters such as the

lead’s resistance and inductance, or the inductance of the ceramic body.

 Chapter 5: MOVs in Variable Frequency Drives 136

(a) [120] (b) [119]

(c) [15] (d) [118]

Figure 5-28. MOV's high frequency models in literature

In models in Figure 5-28(a) and (b) an impedance is added to the model to represent the skin effect.

Skin effect can be modeled with a ladder network of resistors and inductors. In the simplest form it is

considered as a resistor connected in parallel with an inductor, R1 and L1 in Figure 5-28(a) and Zskin in

Figure 5-28(b). Model parameters are determined by numerical algorithms by fitting the measured

impedance to the calculated impedance.

The model shown in Figure 5-28(d), named metal-oxide surge arrester wide-range (MWR) model by

the authors, is intended to provide an accurate model for a wide range of frequencies and amplitudes.

Inductance, L, is determined by experimental formulas and dependent on the height of the arrester and the

number of ZnO columns in parallel. Similarly, the capacitance is considered constant and is calculated by

experimental formula depending on arrester’s height, line discharge class and number of parallel internal

columns of ZnO varistors. RL and Rc are added to the model only for the simulation purpose [118].

The accuracy of the model is mostly dependent on how the non-linear resistor is determined. The

general approach is to measure the voltage of the MOV for different applied currents and draw the
 Chapter 5: MOVs in Variable Frequency Drives 137

measured voltage vs. current curve. Then, a nonlinear curve fitting can determine the relation of voltage

and current of the non-linear element in the model. There are two factor that are affecting the accuracy of

the model. First, obviously, the accuracy of the model depends on how well the assumed function can fit

the measured data. One common function is used in [118] as below:

𝜶
𝒗
𝒊=𝒌 ( ) Equation 5-2
𝑽𝒓𝒆𝒇

where i and v are the current and the voltage of the varistor. α is called the non-linear factor and k is a

constant multiplier. Vref is an arbitrary reference voltage, which normalizes the equation. Although

Equation 5-2 is simple, but it does not fit well on MOV’s I-V curve. To have a better fit, [119] proposes a

different function as below:

𝒗−𝑽𝟎 𝑵
𝒊 = 𝑰𝟎 [𝒍𝒏 (𝟏 + 𝒆 𝑽𝟏 )] Equation 5-3

Where N is an integer number.

Another function that is proposed in [16], used in [15], and is adopted by the manufacturer of the

MOVs used in this thesis, is as following.

−𝐥𝐨𝐠 (𝐈) + 𝐛𝟒 𝐞𝐥𝐨𝐠 (𝐈) ]

𝑽 = 𝟏𝟎 [𝐛𝟏 + 𝐛𝟐 𝐥𝐨𝐠 (𝐈)+ 𝐛𝟑 𝐞 Equation 5-4

The second factor affecting the accuracy of the model is how the voltage and current are measured in

different regions of the curve. As it is described in Chapter 1, an MOV’s I-V curve have 3 regions: leakage

current region, normal clamping region, and up-turn region. Current is in micro- or milli-amperes in the

leakage region, while it can surpass tens of thousands of amperes in normal operation and up-turn regions.

Measuring current and voltage in these regions are conducted with different methods. Typically, in the

leakage current region voltage of the MOV is measured when a given dc current is applied. This is how the

manufacturers create the I-V characteristic curve. For currents around the nominal discharge which can be

several thousand amperes typically 8/20 µs standard surge is used. It should be noticed that the MOV’s

voltage might be different for faster or slower surge waveforms. Figure 5-29 shows the normalized values

for voltage measurement waveforms for a typical MOV [118]. It is seen that the clamping voltage for 4/10

µs waveform is 20% to 30% higher than that for the standard 8/20 µs. In the other end of the curve, in the
 Chapter 5: MOVs in Variable Frequency Drives 138

leakage current region, it is seen that the AC curve do not match the dc curve. The ac leakage current peak

value if much higher than the dc leakage current for low voltages.

Figure 5-29. waveforms used for I-V curve measurement in [118]

Figure 5-30. MOV's leakage current for different frequencies [123]

The AC leakage current itself is dependent on the frequency. The higher the frequency, the higher the

leakage current. An example of the ac leakage current in different frequencies is seen in Figure 5-30. It is

seen that while the dc leakage current is in micro-amperes, the leakage current in 50 Hz is in milli-amperes

and the leakage current for 13 kHz can reach a tenth of an ampere.

Depending on the application, region of interest might be different for modeling. Traditional view is

more concerned about the highest clamping voltage which occurs in the normal operation region. Leakage
 Chapter 5: MOVs in Variable Frequency Drives 139

current region is also important for thermal analysis. However, since the MOV is selected to have a small-

enough leakage current, the models do not need to be accurate in this region. In new applications, most

models are interested in accurately modeling both the leakage current region and normal clamping region.

In an application like VFD protection, the MOV is subjected to repetitive pulses with high frequency and

the thermal stability of the MOV is critical for selecting the proper MOV. Simulating the MOV accurate

enough to be able to calculate its current and power in steady state condition can be helpful in selecting the

proper MOV for VFD systems and non-sinusoidal applications, in general.

5.4.1. Simulation of MOV in VFD system

As it is discussed, the proposed models for the MOV have little difference with each other. The

accuracy of the model mainly relies on the accuracy of modeling the nonlinear resistance. Here, we used

the model shown in Figure 5-28(c). This model is composed of a constant capacitor in parallel with a

nonlinear resistor. A small resistor and a small inductance also are connected in series to model the

resistances and inductances of the leads and connecting wires. For modeling the voltage current curve of

the nonlinear element, the curve given by the manufacturer is used. Previously in Section 5.2, it is shown

that the current and voltage of the MOV matches this curve with an acceptable accuracy.

A nonlinear resistor can be modeled in different ways in the simulation environments. It can be

modeled as a current-controlled voltage source or a voltage-controlled current source. It can also be

modeled as a look up table. Any of these methods might be used in different situations. Here, since the

manufacturer has provided the voltage versus current equation, it is easier to model the nonlinear resistor as

a current-controlled voltage source. It is simulated in MATLAB Simulink and is shown in the figure below.
 Chapter 5: MOVs in Variable Frequency Drives 140

Figure 5-31. MOV model simulated in MATLAB Simulink

Another possibility is to simulate the nonlinear characteristic as a voltage-controlled current source.

Sometimes, especially when the system is big and there are other voltage sources in the simulation,

modeling the MOV as a current source have better solution convergence in simulation. To model an MOV

as a voltage-controlled current source, it is needed to formulate the current in terms of the voltage. Existing

equations 5-2 and 5-3 are not accurate to be used in the model. Therefore, we propose an equation that

matches accurately to the voltage-current curve of the MOV, as it is shown below.

I = 10 ^ ( a1 + a2 × atan ( a3 * log(V) + a4 ) ) Equation 5-5

This equation can be fitted to MOV’s voltage-current characteristic, accurately from micro-amperes to

the nominal discharge current in tens of thousands of amperes. The fitted curve is shown in Figure 5-32.

Figure 5-32. Curve fitting for voltage-controlled current source model

 Chapter 5: MOVs in Variable Frequency Drives 141

The model parameters for current-controlled voltage source model and voltage-controlled current

source model of 150V MOV is given in Table 5-3.

Table 5-3. 150V MOV’s VI curve parameters

Voltage -controlled current source Current-controlled voltage source
I = 10 ^ [ a1 + a2 × atan (a3 × log(V) + a4)] V = 10 ^ [b1 + b2×log(I) + b3×e-log(I) + b4×elog(I)]
a1 -0.8208119 b1 2.4620502
a2 -4.0238658 b2 0.0222331
a3 -9.4237785 b3 -0.0006945
a4 23.0242377 b4 0.0032469

The VFD system simulated in Section 3.5.6 is used here to verify the MOV simulation. Three MOVs

are connected between lines and the ground on the motor terminals. The system is seen in Figure 5-33.

Figure 5-33. VFD system simulation with 150V MOV

Experimental measurement result is shown in Figure 5-26. It is seen that the installing 150V MOVs on

the motor terminals reduces the line to ground overvoltages from about 380 V to about 320 V. In this

condition, the MOVs conduct about 2.5 to 3 amperes of resistive current at the peak of the voltage

waveform.

Simulation results are presented below. Figure 5-34 shows the motor terminal voltage and the MOV

current. It is seen that the MOV conducts 2.7 A current at the peak of the voltage. It is also seen that the

peak value of the line to ground voltage is 320 V.

 Chapter 5: MOVs in Variable Frequency Drives 142

Figure 5-34. Simulation: 150V MOVs on motor terminal, line to ground voltage and MOV current

Simulation allows us to separate the MOV’s resistive and capacitive current, by adding current

measurement inside the MOV model in the resistive and capacitive paths. These currents are shown in

Figure 5-35 alongside the line to ground voltage waveform. From Figure 5-35(a) it is seen that the resistive

current is zero during the pulse, except at the first two voltage peaks.

(a) (b)

Figure 5-35. Simulation: 150 MOV on motor terminal (a) resistive current (b) capacitive current

In summary, the simulation and lab measurements show that the reduction of the high frequency

overvoltages is possible by installing MOVs between lines and grounds. However, the selected MOV

should be able to dissipate the heat generated by the repetitive currents. Also, the capacitive currents of the

MOV, which are drawn from the line, should be considered in the overcurrent protection of the drive.

Furthermore, comparison of the simulation results to lab measurements demonstrate the accuracy and

effectiveness of the simulations.

 Chapter 5: MOVs in Variable Frequency Drives 143

5.5. MOV Application Solutions for VFD systems

VFD systems, like any other electrical system, needs protection against surges. One of the vulnerable

locations can be the cable connecting the inverter to the motor, when it is outdoor. As discussed previously,

the output of the inverter and the motor terminals are recommended locations for SPD installation. The

problem is, the voltage applied to the connecting cable is a PWM voltage. When the cable is long, large

high frequency overvoltages appear on the motor terminal. These overvoltages can discharge large

repetitive currents through the MOV and cause it to fail due to the overheating. Aside from the

overvoltages, it should be noted that MOVs are not tested for PWM voltages and there are no standard

practices for the design engineers to select MOVs for systems with non-sinusoidal voltages such as VFDs.

The common traditional way of SPD design is choosing an MOV with a MCOV higher than the highest

overvoltage in the point of the installation. Considering that the pulse amplitude is usually much higher

than motor’s nominal voltage in an inverter, engineers have to install MOVs with high voltages to avoid

failure. Higher MCOV results in a higher clamping voltage and hence, a compromised protection. For

example, the 220 V motor in the lab is supplied with an inverter with 340 V dc link. To avoid the highest

reflected overvoltages, at least a factor of 2x should be considered. It puts the overvoltage level at 680 V. A

proper MOV installation for this situation would be a 550V MOV with a maximum operating dc voltage of

745 V. With this MOV, for the nominal discharge current of 40 kA, 8/20 µs standard waveform, the

clamping voltage would be higher than 3 kV.

To reduce the protection level to an acceptable range, while avoiding the increased leakage current,

some manufacturers propose a Gas Discharge Tube (GDT) to be installed in series with the MOVs. In this

section, we are going to study this approach and discuss its advantages and disadvantages. Furthermore, we

propose two other solutions for this application: TSPD-Switched MOVs (TSMOV) and Reconfigurable

Surge Protective Device (RSPD). We will investigate the advantages and disadvantages of each method

through theoretical analysis and simulation studies.

5.5.1. MOV in series with GDT

 Chapter 5: MOVs in Variable Frequency Drives 144

GDTs, evolved from traditional Spark Gap, is a commonly used protective device, mainly as the

primary surge protection. It can be used in various surge protection designs with MOVs or TVS diodes. As

mentioned above, one possible solution for MOV application in VFD systems and preventing it from

heating due to the repetitive high leakage currents is adding a GDT in series. In this section, we discuss

different aspects of this solution and compare it to the base solution of installing an MOV with a higher

MCOV, or two MOVs in series. Before that, we briefly review GDTs structure, function, advantages and

disadvantages.

5.5.1.1. Gas Discharge Tube (GDT)

A GDT consists of two electrodes in a sealed tube filled with an inert gas. The distance between the

electrodes, the nature of the gas, and the pressure inside the tube are all controlled to permit a discharge

above a desired voltage. This device is connected in parallel with the protected load, similar to an MOV.

The GDT maintains a high impedance in the OFF state until its voltage exceeds the device’s sparkover or

breakdown voltage. Once the voltage goes beyond this point, the gas molecules become ionized and

conduct electricity. During the discharge, GDT is in low-impedance state and the voltage across it is low

(similar to a short circuit), protecting the load against the incoming surge. Once a GDT is in the arc mode,

it will only stop conducting when the available voltage is below a certain holdover level. In that conditions,

the ions recombine to form molecules and the insulation of the gas is restored. A typical voltage-current

curve of a GDT is shown in Figure 5-36.

Figure 5-36. Typical voltage current characteristic of a Gas Discharge Tube [124]
 Chapter 5: MOVs in Variable Frequency Drives 145

GDTs have low capacitance and leakage current, hence they have minimal effect on normal operation

of the system. On the downside, GDTs are known for their slow response. In other words, the transition

time between high impedance and low impedance states is long (depending on the voltage ramp rate and

amplitude can be up to several tens of microseconds). This might be a serious issue in protecting against

fast transients, like the high frequency overvoltages in PWM inverters. Moreover, GDTs in general have a

short lifespan and can withstand a limited number of surges.

Another major concern about installing the GDTs for the protection of sensitive loads is that the

impulse breakdown voltage (also called sparkover voltage or trigger voltage) is dependent on the voltage

change rate of the pulse. The breakdown voltage response of a GDT to transients with ramp rates of 1 V/µs

or less is referred to as the DC breakdown voltage level. Due to the nature of the gas, the same GDT will

experience breakdown at a higher voltage as the transient’s ramp rate increases. For example, a specific

product with the DC breakdown voltage of 300 V, has a typical value of impulse breakdown voltage of 550

V at 100 V/µs and 800 V at 1 kV/µs [125]. Another disadvantage of the GDT is its inability to stop the

short circuit current after the trigger. The current continues until the voltage decreases to near zero. That is

why they are mainly used in ac systems or in combination with the other devices, such as in series with an

MOV, as mentioned above.

GDTs and MOVs might be connected in different ways. Here, we focus on the series connection of

these two components that might be advertised as a leakage-free SPD. Such a combination is shown in

Figure 5-37. As it is obvious from its name, the obvious advantage is that the leakage current is almost non-

existent. This can protect the MOV from conducting repetitive currents and overheating. Furthermore,

having an MOV in series with the GDT gives it the ability to stop the current after the breakdown.

However, this hybrid products inherits some of the negative aspects of the GDT. Slow transition to the ON

state can make it ineffective in protection against fast transients.

When considering a GDT to be connected in series with an MOV as a hybrid SPD, it is important to

select the breakdown voltage of the GDT so that it does not trigger for the repetitive high frequency

overvoltages. The role of the GDT here is to protect the MOV by not triggering in low repetitive spikes.
 Chapter 5: MOVs in Variable Frequency Drives 146

Figure 5-37. A GDT in series with an MOV

To investigate further, a simple model is developed and simulated in MATLAB Simulink. Figure 5-39

and Figure 5-40 show the model simulated and how it is connected to the surge generator. To simulate the

constant voltage of the GDT in the arc state, a Zener diode is used in the model. A delay module is used in

to represent the intrinsic transition delay of the GDT. The transition delay of a GDT is dependent on the

applied voltage and current and can vary based on the conditions [126]. Figure 5-38 shows the impulse

breakdown voltage of a GDT with DC breakdown voltage of 350 V, surged with generator open circuit

voltages with different peak values. Orange points show the breakdown moments for a GDT in different

tries. Blue curves are generator’s open circuit voltages, plotted to provide a point of reference. It is seen

that when a 400 V, 10/1000 µs surge is applied, the transition delay can be a few tens of microseconds.

When the surge voltage is increased, the transition delay becomes more consistent, but the impulse

breakdown voltage varies more. According to its datasheet, the typical impulse breakdown voltage for this

GDT is 875 V for 1000 V/µs. For higher ramp rates, the impulse breakdown voltage might be even higher.

In the simulations, the breakdown voltage and the transition delay are set to typical values based on the

applied voltage and according to Figure 5-38.

Figure 5-38. GDT transition delay [127]

 Chapter 5: MOVs in Variable Frequency Drives 147

Figure 5-39. Simple GDT model

Figure 5-40. Simulation: surge application to the GDT

The simulation results are shown in Figure 5-41. Three surges are applied to the GDT with surge

generator open circuit voltages of 400, 500 and 1000 volts. For the 400-volt surge, the transition delay can

be long. For the surges below 400 V, GDT does not trigger, even though the DC breakdown voltage is 350

V. For faster surges, this value can be higher. It should be noted that the let-through voltage of a GDT is

dependent on the voltage waveform. This means for the standard 8/20 µs current surge that has a 1.2/50 µs

voltage waveform, the spark is fast, but the let-through voltage of a GDT is much higher than that for an

MOV. As it is seen in Figure 5-42, the maximum let-through voltage of the GDT is 625 V, while this value

for a comparable MOV is 400 V.

(c) OC = 1000 V

(b) OC = 500 V

(a) OC = 400 V

Figure 5-41. 350V GDT’s impulse breakdown voltage and transition delay for different surge
generator voltages
 Chapter 5: MOVs in Variable Frequency Drives 148

(a) (b)

Figure 5-42. 2 kV, 1.2/50 µs voltage surge (1 kA 8/20 µs current) (a) MOV (b) GDT

5.5.1.2. Simulation Results

Assume that the SPD is installed on the motor terminals of a motor, similar to the one in the lab. The

line to ground voltage of the motor is seen in Chapter 3, Figure 3-48. The voltage pulse starts from -170 V

and rises to 170 V, and the oscillations bump the voltage peak to 400 V. The goal is that the designed SPD

should have a low leakage current in 400 V, i.e. GDT should not trigger at the peak voltage. When a GDT

is connected in series with an MOV, it tolerates a large fraction of the voltage in the OFF state, since it is

an open circuit with a much smaller capacitance. Thus, the GDT should be selected to bear all the voltage,

here 400 V. Since the MOV does not tolerate a big voltage in the normal condition, selecting an MOV with

a lower MCOV can achieve an overall lower clamping voltage. However, after the GDT is triggered, the

line voltage is applied to the MOV. The MOV should be able to withstand the full voltage at least until the

next current zero-crossing occurs, where the GDT goes back to the OFF state. Selecting the best MCOV

would be a compromise between the reliability and the performance. Here, it is assumed that the GDT is

connected in series with a 150V MOV. For the comparison with an MOV with higher MCOV, the results

are compared to those from two 150V MOVs connected in series, representing a 300V MOV.

Figure 5-43 shows the simulation of an MOV and a GDT in series. Simulation results show this hybrid

SPD is effective for the slow surges where the GDT response time is not an issue as well as for the large

surges, where the clamping voltage of the MOV can be high.

 Chapter 5: MOVs in Variable Frequency Drives 149

Figure 5-43. Simulation of MOV and GDT in series

(a) (b)

Figure 5-44. 40 kA surge with 1kV/µs ramp, applied to (a) MOV-GDT (b) MOV-MOV in series

(a) (b)

Figure 5-45. 1kA surge with 1kV/µs ramp, applied to (a) MOV-GDT (b) MOV-MOV in series

Figure 5-44 shows the response of the series MOV-GDT for a 40kA surge with a voltage ramp of 1

kV/µs. The response of the GDT depends on the voltage change rate. At this rate, the impulse breakdown

voltage for the GDT is 925 V, based on the datasheet. After the GDT triggers, the voltage drops to MOV

voltage level. The load is subjected to the high voltage only for a very short time, and then voltage drops

almost to half. For the MOV with higher MCOV, or two MOVs in series, the residual voltage is dependent
 Chapter 5: MOVs in Variable Frequency Drives 150

on the amplitude of the current. For this 40kA current surge, voltage surpasses 1500 volts and remains high

for the entire surge duration. For sensitive loads with low thermal capacity, this can have adverse effects

and might cause a failure.

For smaller surges with 1 kV/µs ramp, such as the one in the simulation that is shown in Figure 5-45,

two MOVs in series has a better response. Although the voltage remains around 800 V for all the surge

duration, the voltage peak is smaller than that for the series MOV and GDT. For this surge, GDT triggers in

925 V. After an initial peak, the voltage decreases to about 400 V. This shows that the MOV GDT

combination, exposes the load to higher voltages for the smaller surges. Considering the abundance of the

smaller surges, this can be an important factor in the SPD selection.

The hybrid SPD of an MOV and a GDT in series has other limitations, too. It cannot be customized for

the need by connecting two in parallel. Installing MOVs in parallel is a common practice in industry to

achieve higher surge current handling capability. Even though MOVs do not share the surge current

equally, for large surges this sharing is acceptable in most cases. On the other hand, adding a GDT in series

limits this option. If two GDTs are installed in parallel, once one of them triggers, the voltage will drop and

will prevent the other one from triggering.

5.5.2. TSPD-Switched MOVs (TSMOV)

The Controllable Metal-Oxide Arrester (CMOA) and Thyristor-Switched Arrester have been proposed

in the past for suppression of switching overvoltages in HV and UHV power systems [128], [129]. It is

basically a metal-oxide arrester which is capable of changing the number of varistor blocks by shorting out

a part of the arrester with a power electronic switch in order to maintain a low residual voltage during an

overvoltage. The switch that is used is usually an anti-parallel thyristor which is controlled appropriately by

gate triggering. The voltage at the protected point is monitored, and if an overvoltage is detected, a gate

trigger command is sent to power electronic switch to short a part of the arrester and make the clamping

voltage less. Lower overvoltage level allows to use lower insulation level and reduce the cost of the

equipment.
 Chapter 5: MOVs in Variable Frequency Drives 151

Figure 5-46. TSMOV configuration

The same concept can be used for surge protection purposes, too. The only difference is that the switch

has to be fast enough to act during the surge overvoltage. Based on this concept, we propose to use a

Thyristor Surge Protective Device (TSPD) for reconfiguration of the MOVs in surge application. The

proposed SPD is shown in Figure 5-46. In this configuration, in normal conditions, both MOVs are in the

circuit, providing a higher MCOV to avoid overheating of the MOVs due to overvoltages. When a surge

occurs, MOVs reduce the voltage to some extent. When the voltage across MOV2 reaches the Breakover

voltage of the TSPD, it turns on and shorts out the MOV2, leaving only MOV1 in the circuit. This way, the

clamping voltage seen by the protected load is equal to the voltage of only one of the MOVs. The voltage

of the TSPD in the On state is only a few volts.

Although a TSPD works similar to a GDT, and in some applications, they might be interchangeable,

they have practical differences. For surge protective purposes, TSMOV has several advantages over the

MOV-GDT solution. First, MOVs are connected all the time and the protection is achieved even for small

surges. The second advantage is that TSPD’s switching is more deterministic compared to a GDT. For a

GDT, even though the typical value is given in the datasheet, the impulse breakdown voltage has a random

distribution with a noticeable standard deviation. TSPD, being a semiconductor device, is more predictable

in applications that need a certain accuracy. Another important advantage of the TSPD over a GDT is its

long lifetime. A GDT has a limited number of surges defined as its lifetime, while a TSPD, if properly

designed, has technically unlimited lifetime (service life in excess of 20 years is typical [130]). In this

section, after a brief introduction of the TSPD device, we will discuss the benefits of the proposed method

through simulation results.

 Chapter 5: MOVs in Variable Frequency Drives 152

5.5.2.1. Thyristor Surge Protective Device (TSPD)

Thyristor Surge Protective Device (TSPD) is a bidirectional1 thyristor device designed to protect loads

from lightning and switching over-voltages. The basic thyristor is a NPNP semiconductor device that has

three PN junctions. TSPD is equivalent of a PNP transistor and an NPN transistor connected as a

regenerative pair. It is like a voltage triggered Triac without a gate and operates like a spark gap. It remains

non-conducting until the applied voltage meets or exceeds its rated breakover voltage. Once entering this

conductive state, it continues to conduct, regardless of voltage, until the applied current falls below its rated

holding current. The V-I characteristic of a typical TSPD is shown in Figure 5-47.

Figure 5-47. TSPD voltage current characteristic [130]

Figure 5-48. Switching characteristic of a TSPD

Important parameters of a TSPD are depicted in Figure 5-48. TSPD remains in the Off state as long as

its voltage is below the rated repetitive peak off-state voltage (VDRM). At this voltage, the current is called

repetitive peak off-state current (IDRM). The breakdown region is the high-voltage portion of the voltage-

current characteristic with a low dynamic resistance. Depending on the thyristor design and temperature,

the end of the breakdown region may be at a higher or lower voltage than the start. The maximum voltage

that occurs in the breakdown region is defined as the breakover voltage, VBO. Additional measurements

1
There are various types of TSPDs, including unidirectional TSPDs that are not covered here.
 Chapter 5: MOVs in Variable Frequency Drives 153

may be made of the voltage and current at the switching point (V S, IS). The On-state region is the low-

resistance high-current portion of the voltage-current characteristic. In the On-state condition, TSPD shows

the minimum voltage drop for the current flowing. The minimum current that will just maintain the On-

state condition is defined as the holding current, IH. Currents below this value will cause the thyristor to

switch off.

TSPD is a crowbar protective device with an on-state voltage of a few volts. Generally, a TSPD’s VBO

is 20% to 30% greater than its VDRM. Thus, in this respect it provides better overvoltage protection

compared to a GDT that its typical impulse breakdown voltage can be 150% or more higher than its DC

breakdown voltage. TSPDs have a relatively low energy handling capability and temperature sensitive

characteristics. However, as the majority of the diverted current is conducted in a low-voltage condition,

their current capability is high. The peak impulse let-though voltage will increase as the impulse rate-of-

rise increases. When used within their rated values, these components will have a high life expectancy

[130]. Exceeding device ratings usually results in catastrophic failure.

The junctions, formed by the device NPNP structure, are the principle elements that define the device

capacitance. The capacitance of a PN junction will depend on the applied dc bias, V D, ac test signal level,

Vd, and junction temperature, TJ. At low levels of dc bias, where the ac test level is significant compared to

the dc bias, the capacitance is strongly dependent on the value of ac test level. The effective capacitance

value decreases as the ac test level increases [130].

5.5.2.2. Simulation Results

Similar to the previous section, we consider a VFD system with DC link voltage of 340 V for the

simulations. The SPD is assumed to be installed between lines and ground. The maximum voltage for the

line to ground voltage is 400 V for the system. Since the two MOVs are dividing the voltage, each MOV’s

voltage will not see higher than 200 V. So, the TSPD should not conduct for the normal voltages up to 200

V. A TSPD with the VDRM of 230 V is selected for this situation. The proposed device is simulated in Orcad

Capture software. The PSpice model of the TSPD is provided by the manufacturer. The maximum dynamic

breakover voltage is 295 V for this device, given in the datasheet [131].
 Chapter 5: MOVs in Variable Frequency Drives 154

Figure 5-49. Simulation: applying surge to a TSPD

5.0KA

2.5KA

SEL>>
0A
I(U9:PIN1)
300V

200V

100V

0V
0s 2us 4us 6us 8us 10us 12us 14us 16us 18us 20us
V(U9:PIN1)
Time

Figure 5-50. Simulation: 2/10 µs surge applied to a 230V TSPD and its voltage response

Figure 5-49 shows the ORCAD Capture simulation of the above-mentioned TSPD. A 5 kA, 2/10 µs1

surge is applied to the 230V TSPD. It is seen from the voltage waveform that the maximum voltage is

290V. Also, a GDT-like function is seen from the waveforms. The device is in the non-conducting state for

low voltages. As the voltage reaches the VBO, the device transitions to the low-impedance state and

effectively becomes a short circuit with a low voltage drop.

(a) (b)

Figure 5-51. Simulation: (a) proposed TSMOV (b) Series MOVs for comparison

1
2/10 µs current surge with 2/10 µs open circuit voltage waveform
 Chapter 5: MOVs in Variable Frequency Drives 155

1.0KV 5.0KA 1.0KV 5.0KA

1 2 1 2

0.8KV 4.0KA 0.8KV 4.0KA

0.6KV 3.0KA 0.6KV 3.0KA

0.4KV 2.0KA 0.4KV 2.0KA

0.2KV 1.0KA 0.2KV 1.0KA

>> >>
0V 0A 0V 0A
0s 5us 10us 15us 20us 0s 5us 10us 15us 20us
1 V(MOV2:1) 2 I(Rsurge) 1 V(MOV2:1) 2 I(Rsurge)
Time Time

(a) (b)

Figure 5-52. Simulated surge and voltage response: (a) proposed TSMOV (b) Series MOVs for
comparison

The proposed TSMOV device is seen in Figure 5-51(a). The TSPD is connected in parallel with MOV2.

This MOV will be shorted out during the surge. MOV1 is connected in series and remains in the surge path

all the time. The simulation result is shown in Figure 5-52(a). The simulation result of the 2 MOVs in

series, also, is given in the same figure for comparison. Since the TSPD’s capacitance is much smaller than

the MOV’s capacitance, it does not affect the voltage division between MOVs. TSPD’s capacitance is 65

pF, compared to 4800 pF nominal capacitance of the MOV, has almost no effect. Before TSPD triggers, the

voltage is divided equally among the MOVs (neglecting the MOVs tolerance). Once the TSPD switches to

On state, MOV2 is shorted out and the let-through voltage will be equal to only MOV1’s voltage. For this

example, the maximum let-through voltage is 745 V, but after a sub-microsecond transition time, voltage

drops to around 500 V for the rest of the surge duration. The maximum voltage on MOV2 is the switching

voltage of the TSPD, which is 290V.

For the alternative solution, two MOVs in series, the peak of the let-through voltage is 956 V. This

voltage remains high for the duration of the surge. This is especially important, because generally, the

voltage tolerance capability is higher for shorter overvoltages. For some applications, the let-through

voltage waveform may be just as important as its peak value. They require the length of the overvoltage to

be shorter than a given time, depending on its peak value. Above example shows that TSMOV, not only

reduces the peak value of the let-through voltage, but also reduces the time duration of the highest part of

the overvoltage, considerably.

 Chapter 5: MOVs in Variable Frequency Drives 156

Simulation result shows the effectiveness of the proposed method in reducing the let-through voltage,

without the shortcomings of the GDT-MOV solution. There are also some similarities. Both SPD’s

clamping voltage peak is dependent on the surge ramp rate. However, for TSPD the range of the change in

VBO is smaller than the change in impulse breakdown voltage of the GDT. In summary, the proposed

TSMOV has better performance, longer lifetime, higher reliability than GDT-MOV device. The main

disadvantages are higher price and lower current handling capability of the TSPD, compared to the GDT

based solution. However, being a semiconductor device, TSPDs are expected to have higher surge current

capability and lower price with advancement in fabrication technology.

5.5.3. Reconfigurable Surge Protective Device (RSPD)

Every surge protective device has two states: stand-by state (OFF) and conducting state (ON). In stand-

by state, it is ideal to have zero leakage current and zero capacitance, to have no impact on the normal

function of the system. When a surge occurs, the SPD enters the conducting state, in which the idea is to

prevent the overvoltage and keep the voltage of the system in the same level as before. However, usually a

given parameter has opposite impacts on these two. For example, an MOV with lower MCOV will provide

a lower clamping voltage, but at the same time will have higher leakage current. This is true for two MOVs

in series as well. Installing MOVs in series will decrease the leakage current, but it will increase the let-

through voltage. Considering two MOVs, an ideal connection for the stand-by state is series connection to

reduce the leakage current, while the ideal connection for the conducting state is parallel connection to have

lowered let-through voltage and increased current handling capability.

One way to have two MOVs in series for one state and the same MOVs in parallel for another state is

reconfiguring the SPD. Figure 5-53 shows one such a configuration. This proposed Reconfigurable Surge

Protective Device (RSPD) has two MOVs that can be connected either in series or in parallel.

Reconfiguration is achieved by using three switches in the SPD.

 Chapter 5: MOVs in Variable Frequency Drives 157

Table 5-4. State of the switches for RSPD

State \ Switch S1 S2 S3
MOVs in series OFF OFF ON
MOVs in parallel ON ON OFF

Figure 5-53. Proposed RSPD

When MOV is in stand-by mode, switches S1 and S2 are open and S3 is closed, connecting MOVs in

series between two lines. In occurrence of an overvoltage event, switch S3 closes and switches S1 and S2

close, connecting the MOVs in parallel between the lines. Table 5-4 summarizes the states of the switches

for both stand-by and conducting modes. Figure 5-54 shows the configuration of the SPD in each state.

(a) Stand-by (b) Conducting

Figure 5-54. State of switches for reconfiguration

RSPD provides benefits over other SPDs discussed before. Compared to TSMOV, RSPD can provide

lower let-through voltage, since the current through each MOV is half of that for TSMOV. In TSMOV, the

switched MOV that is shorted out during the surge, experiences less degradation, while the base MOV that

stays in the circuit during the surge degrades more severely. This un-even degradation can change the
 Chapter 5: MOVs in Variable Frequency Drives 158

voltage division ratio for the MOVs. In RSPD both MOVs experience similar surges and system voltages,

thus their degradation levels are more likely to be close to each other.

Decision to turn on or off the switches can be made based on the voltage amplitude of the SPD, which

is equal to the load voltage. Depending on the maximum desired voltage on the load, a simple comparison-

based circuit can be used to trigger the switches.

In order to avoid creating a voltage pulse due to disconnecting the surge current, switches should be

triggered in order. In normal operating condition, S3 is closed (ON), and the other two switches are open

(OFF). In the advent of the surge, S3 will carry the surge current, when the surge is still small. Once the

voltage reaches the threshold value, S1 triggers first short-circuiting MOV2 and S3. After S1 closes, S3

opens. After that, S2 closes, providing the second surge current path through MOV2, in parallel with

MOV1.

5.5.3.1. Simulation Results with Ideal Switches

Figure 5-55 shows simulation of the concept of the RSPD in Matlab Simulink. 150 V MOV model built

in previous sections is used here, with ideal switches. Trigger decision of the switches are made based on

the load voltage, seen as “Reconfig Ctrl” unit in the figure. This unit, compares the load voltage to a

reference value, here 400 V. When voltage reaches the threshold value, it provides command signals to the

gates of the switches with a small delay, as it is discussed above.

Figure 5-55. Simulation of RSPD with ideal switches

 Chapter 5: MOVs in Variable Frequency Drives 159

(a) (b)

Figure 5-56. Simulation results for (a) RSPD, compared to (b) two MOVs in series

Figure 5-57. Simulation results, switching moment of the MOVs at 400V from series to parallel

Simulation results demonstrate the effectiveness of the RSPD. In Figure 5-56, let-through voltage of the

proposed RSDP is compared with that of two MOVs in series. In normal conditions of the system, both of

these two configurations would have low leakage current, since both have two MOVs in series. Under

surge condition, two MOVs in series has higher let-through voltage. As it is seen from the figure, for a 5

kA 8/20 µs standard surge, the let-through voltage reaches 1000 volts. For the same surge, RSPD provides

the protection with the let-through voltage of 453 volts. This is a substantial improvement by utilizing

reconfiguration of the existing MOVs under surge condition.

Figure 5-57 shows the voltage when MOVs are switching from series to parallel connection. The

threshold is set to trigger when voltage reaches 400 V. When MOVs is shorted out by S1, voltage drops to

half of 400V, i.e. 200V instantly. Then, after a short delay, S2 reconnects MOV2, this time in parallel with
 Chapter 5: MOVs in Variable Frequency Drives 160

MOV1. MOVs share the surge current, so their voltages decrease further. As it is seen from the figure,

voltage decreases to 160V, and then increased again according to MOVs current-voltage characteristic.

Summary of the results is presented in Table 5-5.

Table 5-5. Let-through voltage reduction by RSPD for 8/20 µs surge

Surge current Max voltage (V)
Improvement (%)
peak (kA) RSPD Two MOVs in series
5 453 1000 54.7
10 502 1142 56.0
20 576 1364 57.7
40 692 1725 59.8

5.5.3.2. Switches in RSPD

Switches in RSPD can be discussed in two aspects: switch type and its surge current handling

capability. Since these switches (two of them: S1 and S2) are in the surge current path, they need to

conduct the surge current without considerable voltage drop on them. They also need to have high surge

current handling capability. Another important factor is the switching delay. The overall reaction time from

the advent of the surge until having the first MOV on should be acceptably short. A part of this reaction

time is the control circuit delay, which in case of using a simple analog comparison circuit is short. Thus,

the turn on delay of the switch should be short enough to be able to turn on before surge voltage reaches a

damaging level.

Considering the requirements of the switches in the RSPD, insulated-gate bipolar transistor (IGBT)

seems to be a good choice. An IGBT is a three terminal, four-layer semiconductor device (PNPN) that

works as a unidirectional switch. In a simple view, an IGBT is a combination of a MOSFET and a bipolar

junction transistor (BJT) to achieve high current handling capability of BJTs while maintaining the simple

gate drive requirements of MOSFETs. Figure 5-58 shows IGBT’s cross-section and its internal connection

as a MOSFET and a BJT, as well as its circuit symbol.

 Chapter 5: MOVs in Variable Frequency Drives 161

Figure 5-58. IGBT: insulated-gate bipolar transistor

The main advantages of an IGBT over a Power MOSFET and a BJT are:

1. It has a low on-state voltage drop and superior on-state current capability.

2. Low driving power and a simple drive circuit due to the input MOS gate structure.

3. It has high reverse blocking capabilities.

The main drawbacks of IGBTs are:

1. Switching speed is inferior to that of a Power MOSFET and superior to that of a BJT. The

collector current tailing causes the turnoff speed to be slow.

2. IGBT only conducts in one direction. It does not have a body diode, and usually a reverse

conducting diode should be added to the module.

Early generations of the IGBT also had other issues, such as being prone to latch up or having a

negative temperature coefficient, which could lead to thermal runaway and makes the paralleling of devices

difficult. However, nowadays these issues are addressed and resolved in present generations of IGBTs

[132].

One aspect of selecting a switch for a surge protection application is the fact that surges can occur in

positive or negative polarity. Thus, the switch should have the capability of conducting the current in both

directions. An IGBT cannot conduct current in the reverse direction (from emitter to collector). This is

unlike a power MOSFET, which can conduct currents in either direction (drain to source or source to

drain). Also, unlike MOSFET, an IGBT does not have an intrinsic body diode that allows the current to

flow in reverse direction, when the device is off. Therefore, in most applications, a fast recovery diode is

connected in anti-parallel with the IGBT, as shown in Figure 5-59(a).

 Chapter 5: MOVs in Variable Frequency Drives 162

(a) (b)

Figure 5-59. IGBT (a) anti-parallel diode (b) bidirectional conducting

Another aspect of selecting switch for surge protection application is its blocking characteristic. The

switch should be able to block in both directions. Although an IGBT has capability of blocking forward and

reverse polarity voltages, IGBTs with anti-parallel diodes cannot block the reverse polarity voltages.

Therefore, to achieve both bidirectional conducting and bidirectional blocking, it is needed to use two

IGBTs in series, as it is shown in Figure 5-59(b). When surge flows through a bidirectional switch of this

type (two IGBTs in series), the total voltage drop on the device would be sum of the forward conducting

voltage of one IGBT and forward conducting voltage of one anti-parallel diode. For an acceptable switch

for RSPD, this voltage should be considerably less than MOV’s voltage during the surge.

Behavior of electronic switches, among them IGBTs, are studied in literature [133]–[139]. However, in

most cases, these papers investigated short circuit current handling capability and do not focus on lightning

surges. Since electronic switches are not mainly designed for conducting lightning surge currents, little

information is available on their surge current handling capability using 8/20 µs or other standard

waveforms in microsecond ranges.

Reference [135] tested IGBT’s capability in conducting 10 ms half sine shape surge current pulse and

shows that with a higher gate voltage, it can conduct 20 times the rated current without failure. If we can

assume the same I2t for a 20 µs surge current, the switch would be able to handle 8/20 µs surges of 220

times its rated current. For an IGBT with 100 A rated current, the surge handling capability would succeed

20 kA. If the switch is optimized for surge current handling, it may be able to conduct even higher currents.

Figure 5-60 shows IGBT’s voltage during the half sine wave current surge from [135]. IGBT’s voltage

peak reaches 20 V for a current 15 times its rated current.

 Chapter 5: MOVs in Variable Frequency Drives 163

(a) (b)

Figure 5-60. IGBT characteristics during the surge (a) Voltage (b) IC-VCE [135]

As it is stated previously, in a bidirectional IGBT, shown in Figure 5-59, the fast recovery diode’s

voltage, also, needs to be small for the surges. Nowadays, fast recovery diodes can conduct high surge

currents with a small voltage drop. Figure 5-61 shows forward voltage of the IGBT’s anti-parallel diode

during an 80 kA, 100 µs surge current, for a 1700A IGBT module. As it is seen, the peak voltage is less

than 20 V during the peak voltage. This high surge current handline capability is achieved through well

balanced parallelization of diode inside the module. Most of the switches are designed to have positive

temperature coefficient at rated current and above, resulting in homogeneous current flow through diodes.

Figure 5-61. Fast recovery diode: forward voltage vs surge current (80 kA, 100 µs) [133]

Although there is still little published on surge current capability of electronic switches, above examples

show that it is possible to achieve high surge current handling capability that can be used in RSPD.
 Chapter 5: MOVs in Variable Frequency Drives 164

5.6. Summary and Conclusions

Generally, MOVs are designed to conduct occasional surges in the power systems with sinusoidal

voltage waveform. However, in new electrical environment, sometimes engineers need to install surge

protection devices for a system with non-conventional voltage. VFD systems, as an example, use PWM

voltage over a cable connecting the inverter to the motor that needs to be protected against surges. In such

conditions, repetitive overvoltages created on motor terminal can conduct large repetitive currents through

MOV and cause its failure. In this chapter, we studied MOV’s behavior under repetitive surges from a

voltage supply as well as PWM pulses. It is discussed how installing MOV on a VFD system can affect the

wave shape of overvoltages. Also, we studied MOV’s high frequency model and simulated the MOV used

in the tests and compared the simulated waveforms to the measured ones from the lab. Furthermore, we

demonstrated MOVs can reduce the high frequency overvoltages in VFD systems. However, in this

condition MOVs conduct resistive currents that can raise the MOVs temperature. Further investigation of

this application is suggested as a potential future research.

Furthermore, in this chapter, we proposed two new approaches for installing MOVs on PWM lines to

prevent MOV failure due to high leakage current, while maintaining low protection level during the surges.

These two methods, TSPD-Switched MOVs (TSMOV) and Reconfigurable Surge Protective Device

(RSPD) are simulated and compared to existing methods of installing two MOVs in series or connecting a

GDT in series with MOVs. Through simulation studies, effectiveness of the proposed methods is shown,

and their advantages and shortcomings are discussed in detail. Comparisons show the proposed method

have superior performance over the existing methods.

 Chapter 6: Conclusions and Future Work 165

Chapter 6:

6. Conclusions and Future Work

The research in this thesis focuses on the topic of surge protection. New methods for the lifetime

estimation of Metal Oxide Varistors (MOVs) are proposed, and new approaches for surge protection of

Variable Frequency Drives (VFDs) are presented and experimentally validated. The thesis first a thorough

review the failure modes of MOVs and the influential factors for their degradation are explained. Then

multiple experiments with a large number of MOVs are conducted to better model and explain the Energy

Absorption Capability (EAC) of both new and degraded low voltage varistors. This represents the first

extensive study on the impact of EAC degradation for low voltage MOVs. Furthermore, a new method is

proposed to estimate the remaining lifetime of the MOVs.

The rest of the thesis focuses on the surge protection of the VFD systems. With the increasing

importance of the energy efficiency in today’s power system, VFD applications are expanding rapidly. As a

main issue in their surge protection, this thesis investigates, in detail, the problem of high frequency

overvoltages that are created on the motor terminal due to the voltage reflection phenomenon in long

cables. The MOV is typically built for conducting occasional surge currents, where it has enough time to

cool down after a surge event. The high frequency overvoltages can conduct high repetitive currents

through the MOV and cause its failure because the cool down period is eliminated. A comprehensive
 Chapter 6: Conclusions and Future Work 166

literature review is carried out on the causes, effects and the mitigation of these overvoltages. Qualitative

explanations of the failures, supported by analytic models, are presented.

Besides the theoretical work, an experimental VFD system is built in the lab at Northeastern University.

This setup is comprised of an induction motor controlled by a VFD. Motor shaft is coupled to a

dynamometer as a mechanical load. As the controller, a programmable motor control development board is

used. The cable and the motor models are extracted for the experimental setup, and the system is simulated.

Moreover, MOVs behavior are studied under pulse voltages. At the end, two new approaches are proposed

for installing MOVs safely on the VFD systems, that provide superior performance over the existing

methods. In this chapter, the results and the conclusions of this thesis are summarized, and some novel

ideas are suggested for the future work.

6.1. Summary of Results

The main research contributions of this thesis are summarized below, and are repeated from previous

chapters:

In Chapter 1, a literature review is conducted on the failure modes of the MOVs. Thermal runaway,

puncture, cracking and flashover are identified as main failure modes of the MOVs. The effective factors

are the surge magnitude and duration, MOV’s microstructural non-uniformity, size, coating and contacts.

In Chapter 2, experimental results on new and degraded low voltage MOVs are presented, and their

EAC and Time to Failure (TtF) are analyzed. These experiments showed that average EAC of MOVs

decrease with surge degradation. It is also found that unipolar surges decrease EAC more than bipolar

surges. Also, the results reveal that degraded MOVs conduct lower currents during a transient overvoltage

(TOV). This fact leads to the conclusion that surge degraded MOVs have longer average TtF than new

MOVs for a given TOV. Finally, the experimental tests performed on different MOV’s showed that some

commercial products might have very similar nominal specifications, yet they have a significant difference

in their surge capabilities. In addition to the test results, a new health monitoring algorithm for low voltage

MOVs is proposed. The proposed algorithm utilizes takes the MOV’s voltage and current waveforms and

its temperature as inputs and then estimates its remaining lifetime.

 Chapter 6: Conclusions and Future Work 167

Chapter 3 describes the voltage reflection phenomenon in a VFD system and discusses the influential

factors on overvoltage in VFD. It is shown that overvoltage values are mainly dependent on the pulse rise

time and the cable length. Mathematical calculations demonstrate that if the voltage pulse rise time is

shorter than twice the traveling time, a full reflection will occur. In such cases, the motor terminal voltage

will reach almost double the voltage of the DC link. Furthermore, in this chapter, an experimental VFD

system is used to validate the analytic analysis. The detailed models for the cable and the motor are

extracted from the lab test-bench, and the system is simulated. Simulation results demonstrated that not

always increasing the rise time can lead to the decreased overvoltage, because of the second and subsequent

voltage wave reflections. The actual relation is presented through simulations.

Chapter 4 presents a thorough literature review on the overvoltage mitigation techniques in VFD

systems, and advantages and disadvantages of each method is discussed. The main approach to mitigate the

overvoltages in the VFD systems is using passive or active filters. These filters can be installed on the

motor terminals or the inverter output, or even both. Other mitigation techniques include PWM control,

using a dc cable between the rectifier and inverter instead of an ac cable between inverter and motor, and

installing high energy varistors on the motor terminal.

Chapter 5 reports the experimental findings on MOV behavior under repetitive pulses. It is shown that

installing MOVs at the motor terminal can reduce the overvoltages created by voltage wave reflection. It is

also shown that the resistive current during a pulse can behave in different ways. It can always be

increasing, or it can increase in the beginning of the response and then decrease. The response depends on

on several factors, including inductance of the ceramic, the waveform of the voltage pulse, MOV’s

microstructure, and the heating of the MOV. Conducting resistive current heats up the MOV. Thus, it is

necessary to make sure the average power loss does not exceed the power dissipation capability of the

MOV. For such applications, designers need to know the highest current that an MOV can draw for a

specific voltage. So, it is suggested that the manufacturers provide the lower limits of voltage-current

characteristic.

Furthermore, two new approaches, TSPD-Switched MOVs (TSMOV) and Reconfigurable Surge

Protective Device (RSPD), are proposed in Chapter 5. The proposed methods prevent MOV failure due to
 Chapter 6: Conclusions and Future Work 168

high leakage currents, while maintaining low protection level during the surges. Effectiveness of the

proposed methods are demonstrated through simulation studies. The results show the proposed methods

have superior performances over the existing methods.

TSMOV is comprised of two MOVs in series, where one of them is connected in parallel with a TSPD.

During the normal condition, the two MOVs are connected in series, reducing the leakage current. Under

the surge condition, TSPD is shorting out one of the MOVs, providing lower clamping voltage. In

summary, the proposed TSMOV has better performance, longer lifetime, higher reliability compared to the

existing method of series GDT-MOV device.

Similar to the TSMOV, RSPD has two MOVs, where it can reconfigure them into parallel or series

connection. This allows MOVs to be connected in series under normal voltage, and in parallel during the

surge occurrences. The possibility and effectiveness of the RSPD is demonstrated through simulation.

6.2. Suggestions for Future research

Below are a few ideas for a possible future research that might build upon the research of this thesis.

• Near-failure MOV detection using machine learning

MOVs are known to degrade over time when they experience surges and transient overvoltages.

Usually, to avoid fire hazard and damages to the system, they are thermally protected. This means that once

they reach their end of life, they fail and get disconnected from the line, leaving the load unprotected. For

sensitive systems, the traditional way of monitoring the health of MOVs are periodic assessment of all the

installed MOVs by disconnecting them and measuring their V1mA. This costly operation can be replaced by

an initial investment on the electronic monitoring system that identifies the near-failure MOVs. This goal

might be achieved using Machine Learning and artificial Neural Network.

Artificial Neural Network (ANN) is an imitation of biological neural network in the brain. It is a system

of nodes and neurons, that receive inputs and send outputs to each other. While in nature, inputs and

outputs of the neurons are electrical signals, in the ANN, numerical values are used. Outputs of nodes are

determined by calculating the weighted sum of the inputs to represent the total strength of the input signals,
 Chapter 6: Conclusions and Future Work 169

adding a bias value and applying a nonlinear activation function on the sum. These outputs are then fed to

the other layers as inputs.

Figure 6-1. A typical Deep Neural Network

A typical ANN is shown in Figure 6-3. Each ANN consist of input layer, hidden layers and output

layers. The neural network with more than one hidden layer is called Deep Neural Network (DNN). The

input layer consists of a list of input features. There are various types of layers, including fully connected,

convolution and pooling that allow various types of mathematical operations. Each neuron, shown as a blue

line in the figure, represent a weight.

Depending on the problem and data set, the ANN can be trained in two ways: supervised or

unsupervised learning. For the supervised learning, a labeled data set is needed. Using unsupervised

learning the data set will be unlabeled and the network will learn by itself. Training an ANN means finding

the appropriate weights of the neural connections.

In the problem of near-failure MOV detection, the inputs are components of currents and voltages of the

MOV such as amplitude, frequency, and harmonics. Temperature of the MOV’s surface could also be used

as an input. A large set of offline measurements of healthy and degraded MOVs are needed to train the

ANN. The larger the training data set, the more accurate the results, so future research would emphasize

performing more extensive experimental MOV measurements than performed in this thesis. The data set

should contain the MOVs voltage and current information under various voltages within the acceptable

voltage range (normally -10% to +5% of nominal value) in different temperatures (for example -20 °C to

85 °C), and whether this MOV is needed to be replaced, based on the health criteria. For example, if V 1mA
 Chapter 6: Conclusions and Future Work 170

is below 90% of its nominal value the MOV is labeled failed. During inference, data from online

measurement would be fed to the network, and the output of the network would determine if this MOV

needs to be replaced. This proposed concept is depicted in Figure 6-2. Fundamental components of voltage

and current (V1, I1), their harmonics (V3, I3, …), and temperature (T) is measured and fed to the network

as inputs. However, future research should also investigate whether other signals should be monitored and

measured for the accuracy of the neural network. The goal would be for the network to provide a simple

pass/fail output for the health of MOV.

Figure 6-2. Near-failure MOV detection with ANN

• Evaluation of effects of PWM pulses on lifetime of the MOVs

With the expansion of the power electronics in power systems, more MOVs are needed to be installed

on the lines with non-sinusoidal voltages. Manufacturers provide data on the life expectancy of the MOVs,

based on the conventional sinusoidal voltage tests. Accelerated aging tests can be designed to study the

effect of PWM voltage on the lifetime of the MOVs.

The following experiments are suggested in this regard:

The peak resistive leakage current and the power loss of new MOVs could be measured, as the initial

states. Then MOVs should be installed in a VFD system, on the motor terminal for a given period. The test

can be conducted in an enclosed controllable oven to allow control of the environment temperature. After

the exposure period, the peak resistive current and power loss of MOVs can be measured again. These data

can be compared to the data collected from a control group that were under MCOV with sinusoidal
 Chapter 6: Conclusions and Future Work 171

waveform for a similar period of time. An excess increase of the power loss would indicate a faster

degradation. Multiple test groups are needed to be tested under different DC link voltages and switching

frequencies. A challenging obstacle of this type of research is that it requires statistical study and multiple

samples should be used to compensate for the randomness of the MOV’s behavior.

• Mitigation of the high frequency overvoltages in VFD using combination of MOVs and

filters

There have been extensive studies on how to design a filter for mitigation of overvoltages on motor

terminal due to the voltage wave reflection. Typically, the filters have capacitors connected between lines

and a neural point. However, future research might attempt to utilize the MOV’s capacitance to reduce the

size of the filter capacitors, and at the same time, allow the MOV to be installed on the line to provide the

surge protection. This can resolve the installation problem for the MOV with an added benefit of having a

smaller capacitance for the filter. In fact, it might lead to different types of MOVs to be designed to have

increased filtering abilities specific for the VFD overvoltage mitigation.

Furthers, traditional filters in VFDs can either be passive or active filters. It would be interesting

research to include the effects of the MOVs directly into the topology design of the filters. One possible

topology is suggested in Figure 6-3. The MOV is connected in parallel to the filter capacitor. Filter

parameters will be calculated, and perhaps even reduced, considering the MOV’s capacitance. This

topology allows MOVs to be installed between a line and a neutral point. The impact on lifetime and

performance of the combined MOV and passive filter design should be compared with the traditional

passive filter designs now presently utilized.

Figure 6-3. Combination of MOV and passive filter

 Chapter 6: Conclusions and Future Work 172

Further, it may be possible to incorporate MOVs in the active filters as well. The main requirement is

that MOVs should be directly connected between lines and neutral (or ground). Resistors or inductors

should be avoided in series with MOVs, because they increase the impedance of the surge path and make

the surge protection ineffective. This would introduce an entire new area of research in VFD protection.
 References 173

Bibliography

[1] “Surge Protection Reference Guide,” Emerson Network Power. 2011.

[2] B. S. Reddy and A. R. Verma, “Protection of Digital Telecom Exchanges Against Lightning Surges and Earth Faults,” IEEE
Trans. Ind. Appl., vol. 51, no. 6, pp. 5305–5311, 2015.
[3] T. Miyazaki, T. Ishii, and S. Okabe, “A Field Study of Lightning Surges Propagating Into Residences,” IEEE Trans.
Electromagn. Compat., vol. 52, no. 4, pp. 921–928, 2010.
[4] R. J. Campbell, “Weather-Related Power Outages and Electric System Resiliency,” Congressional Research Service Report
for Congress. 2012.
[5] C. F. Barbosa and F. E. Nallin, “Lightning Protection of a Smart Grid Sensor,” in 2013 International Symposium on
Lightning Protection (XII SIPDA), 2013, pp. 273–277.
[6] D. Hill, “Why Power Supplies Fail,” in 2016 IEEE Applied Power Electronics Conference and Exposition (APEC), 2016.
[7] S. L. Capitaneanu, B. De Fornel, M. Fadel, F. Jadot, and F. Malrait, “Modelling and identification for reduction of
overvoltage transients,” IEE Proc. - Electr. Power Appl., vol. 150, no. 2, p. 175, 2003.
[8] S. Amarir and K. Al-Haddad, “A modeling technique to analyze the impact of inverter supply voltage and cable length on
industrial motor-drives,” IEEE Trans. Power Electron., vol. 23, no. 2, pp. 753–762, 2008.
[9] “IEEE Std C62.41.1TM-2002, IEEE Guide on the Surge Environment in Low-Voltage (1000 V and Less) AC Power
Circuits.” 2002.
[10] “ITI ( CBEMA ) CURVE APPLICATION NOTE,” Information Technology Industry Council (ITI), 2000. [Online].
Available: www.keysight.com/upload/cmc_upload/All/1.pdf.
[11] “National Lightning Detection Network, Density Maps.” [Online]. Available:
http://www.vaisala.com/en/weather/lightning/Pages/default.aspx.
[12] O. M. Clark and R. E. Gavender, “Lightning protection for Microprocessor-Based Electronic Systems,” IEEE Trans. Ind.
Appl., vol. 26, no. 5, pp. 947–953, 1990.
[13] “IEEE Std C62.41.2TM-2002, IEEE Recommended Practice on Characterization of Surges in Low-Voltage (1000 V and
Less) AC Power Circuits.” 2002.
[14] R. A. Jones et al., “Modeling of Metal Oxide Surge Arresters,” IEEE Trans. Power Deliv., vol. 7, no. 1, pp. 302–309, 1992.
[15] C. Mihalache, S. Nitu, and F. Ionescu, “Varistor modeling for power electronics applications simulations,” J. Optoelectron.
Adv. Mater., vol. 10, no. 5, pp. 1272–1276, 2008.
[16] M. Holzer and W. Zapsky, “Modeling varistors with PSpice: Simulation beats trial and error,” Siemens Components-English
Edition, vol. 32, no. 1. pp. 5–7, 1997.
[17] M. Bartkowiak, M. G. Comber, and G. D. Mahan, “Failure modes and energy absorption capability of ZnO varistors,” IEEE
Trans. Power Deliv., vol. 14, no. 1, pp. 152–160, 1999.
[18] J. He and J. Hu, “Discussions on nonuniformity of energy absorption capabilities of ZnO varistors,” IEEE Trans. Power
Deliv., vol. 22, no. 3, pp. 1523–1532, 2007.
[19] K. Eda, “Destruction mechanism of ZnO varistors due to high currents,” J. Appl. Phys., vol. 56, no. 10, pp. 2948–2955,
1984.
[20] A. Vojta and D. R. Clarke, “Microstructural origin of current localization and ‘“puncture”’ failure in varistor ceramics,” J.
Appl. Phys., vol. 81, no. 2, pp. 985–993, 1997.
[21] M. Kan, S. Nishiwaki, T. Sato, S. Kojima, and S. Yanabu, “Surge Discharge Capability and Thermal Stability of a Metal
Oxide Surge Arrester,” IEEE Trans. Power Appar. Syst., vol. PAS-102, no. 2, pp. 282–289, 1983.
[22] S. Tominaga, Y. Shibuya, Y. Fujiwara, M. Imataki, and T. Nitta, “Stability and Long Term Degradation of Metal Oxide
Surge Arresters,” IEEE Trans. Power Appar. Syst., vol. PAS-99, no. 4, pp. 1548–1556, Jul. 1980.
[23] J. He, J. Liu, J. Hu, R. Zeng, and W. Long, “Non-uniform ageing behavior of individual grain boundaries in ZnO varistor
ceramics,” J. Eur. Ceram. Soc., vol. 31, no. 8, pp. 1451–1456, 2011.
 References 174

[24] E. Olsson and G. L. Dunlop, “Characterization of individual interfacial barriers in a ZnO varistor material,” J. Appl. Phys.,
vol. 66, no. 8, pp. 3666–3675, 1989.
[25] Z. Nanfa, K. Guoyao, and G. Yaping, “Long duration impulse withstand capability of SPD,” in 2010 Asia-Pacific
Symposium on Electromagnetic Compatibility, APEMC 2010, 2010, pp. 1510–1513.
[26] N. Tsukamoto, “Surge Withstand Capability of Metal Oxide Varistors for 10 / 350 us Waveform,” in 2011 International
Symposium on Lightning Protection (XI SIPDA), 2011, pp. 177–181.
[27] A. Paul, S. Saravanan, N. Jha, J. Yesuraj, and J. B. Nemade, “Evaluation and failure analysis of high-resistance collar
coating of varistor,” in Proceedings of the IEEE International Conference on Properties and Applications of Dielectric
Materials, 2012, pp. 2–5.
[28] M. Darveniza and T. K. Saha, “Surface flashovers on metal-oxide varistor blocks,” in IEEE International Conference on
Conduction & Breakdown in Solid Dielectrics, 1998, pp. 406–409.
[29] M. Darveniza, L. R. Tumma, B. Richter, and D. A. Roby, “Multipulse lightning currents and metal-Oxide arresters,” IEEE
Trans. Power Deliv., vol. 12, no. 3, pp. 1168–1172, 1997.
[30] K. G. Ringler, P. Kirkby, C. C. Erven, M. V. Lat, and T. A. Malkiewicz, “The energy absorption capability and time-to-
failure of varistors used in station-class metal-oxide surge arresters,” IEEE Trans. Power Deliv., vol. 12, no. 1, pp. 203–208,
1997.
[31] M. M. R. Ahmed, G. A. Putrus, L. Ran, and R. Penlington, “Measuring the energy handling capability of metal oxide
varistors,” in IEE Conference Publication, 2001, vol. 1, no. 482.
[32] H. Imagawa et al., “PD signal propagation characteristics in GIS and its location system by frequency components
comparison,” IEEE Trans. Power Deliv., vol. 16, no. 4, pp. 564–570, 2001.
[33] M. Bartkowiak, G. D. Mahan, F. A. Modine, M. A. Alim, R. Lauf, and A. McMillan, “Voronoi network model of ZnO
varistors with different types of grain boundaries,” J. Appl. Phys., vol. 80, no. 11, pp. 6516–6522, 1996.
[34] J. He, J. Hu, and Q. Chen, “Microstructure Simulation on Puncturing Phenomenon of ZnO Varistor under High Current,” in
2006 IEEE 8th International Conference on Properties and applications of Dielectric Materials, 2006, no. April, pp. 959–
962.
[35] Y. Tu, L. Ding, J. He, J. Hu, and R. Zeng, “Effect of nonuniformities of microstructure and electrical property of grain
boundary to the global electrical characteristics,” Proc. IEEE Int. Conf. Prop. Appl. Dielectr. Mater., pp. 95–98, 2007.
[36] D. Q. Tan, K. Younsi, Y. Zhou, P. Irwin, and Y. Cao, “Nano-enabled metal oxide varistors,” IEEE Trans. Dielectr. Electr.
Insul., vol. 16, no. 4, pp. 934–939, 2009.
[37] M. Bartkowiak, M. G. Comber, and G. D. Mahan, “Energy handling capability of ZnO varistors,” J. Appl. Phys., vol. 79, no.
11, p. 8629, 1996.
[38] A. N. M. Karim, S. Begum, and M. S. J. Hashmi, “Performance and failure during energy testing of zinc oxide varistor
processed from different powder size fraction and passivation thickness,” Int. J. Mech. Mater. Eng., vol. 5, no. 2, pp. 175–
181, 2010.
[39] S. Boggs, J. Kuang, H. Andoh, and S. Nishiwaki, “Increased energy absorption in ZnO arrester elements through control of
electrode edge margin,” IEEE Trans. Power Deliv., vol. 15, no. 2, pp. 562–568, 2000.
[40] D. Talebi Khanmiri, R. Ball, and B. Lehman, “Degradation Effects on Energy Absorption Capability and Time to Failure of
Low Voltage Metal Oxide Varistors,” IEEE Trans. Power Deliv., vol. 32, no. 5, pp. 2272–2280, 2017.
[41] D. Talebi Khanmiri, R. Ball, and B. Lehman, “Unipolar and bipolar degradation of low voltage Metal Oxide Varistors,” in
2016 IEEE Power and Energy Society General Meeting (PESGM), 2016, pp. 1–5.
[42] D. Talebi Khanmiri, R. Ball, C. McKenzie, and B. Lehman, “Energy absorption capability of low voltage Metal Oxide
Varistors in AC and impulse currents,” in 2016 IEEE Applied Power Electronics Conference and Exposition (APEC), 2016,
pp. 3038–3042.
[43] D. Talebi Khanmiri, R. Ball, J. Mosesian, and B. Lehman, “Degradation of Low Voltage Metal Oxide Varistors in Power
Supplies,” in 2016 IEEE Applied Power Electronics Conference and Exposition (APEC), 2016, pp. 2122–2126.
[44] M. N. Tuczek and V. Hinrichsen, “Recent experimental findings on the single and multi-impulse energy handling capability
of metal-oxide varistors for use in high-voltage surge arresters,” IEEE Trans. Power Deliv., vol. 29, no. 5, pp. 2197–2205,
 References 175

2014.
[45] E. C. Saksshaug, J. J. Burke, and J. S. Kresge, “Metal Oxide Arresters on Distribution Systems Fundamental
Considerations,” IEEE Trans. Power Deliv., vol. 4, no. 4, pp. 2076–2089, 1989.
[46] C. De Salles, M. L. B. Martinez, and Á. A. A. de Queiroz, “Surge Ageing of Metal Oxide Varistors,” in 2015 Electrical
Insulation Conference (EIC), 2015, no. June.
[47] M. A. Ramírez et al., “The failure analyses on ZnO varistors used in high tension devices,” J. Mater. Sci., vol. 40, no. 21,
pp. 5591–5596, 2005.
[48] K. P. Mardira, T. K. Saha, and R. A. Sutton, “Investigation of diagnostic techniques for metal oxide surge arresters,” IEEE
Trans. Dielectr. Electr. Insul., vol. 12, no. 1, pp. 50–59, 2005.
[49] C. Heinrich and V. Hinrichsen, “Diagnostics and monitoring of metal-oxide surge arresters in high-voltage networks-
comparison of existing and newly developed procedures,” IEEE Trans. Power Deliv., vol. 16, no. 1, pp. 138–143, 2001.
[50] “MO Surge Arresters Stresses and Test Procedures,” Cigre TB 544, no. August. pp. 1–149, 2013.
[51] V. Hinrichsen, “Monitoring of High Voltage Metal Oxide Surge Arresters,” in VI Jornadas Internacionales de Aislamiento
Eléctrico, 1997.
[52] Z. Xu, L. Zhao, A. Ding, and F. Lü, “A Current Orthogonality Method to Extract Resistive Leakage Current of MOSA,”
IEEE Trans. Power Deliv., vol. 28, no. 1, pp. 93–101, 2013.
[53] Y. Han, Z. Li, H. Zheng, and W. Guo, “A Decomposition Method for the Total Leakage Current of MOA Based on Multiple
Linear Regression,” IEEE Trans. Power Deliv., vol. 31, no. 4, pp. 1422–1428, 2016.
[54] M. Khodsuz and M. Mirzaie, “Evaluation of ultraviolet ageing , pollution and varistor degradation effects on harmonic
contents of surge arrester leakage current,” IET Sci. Meas. Technol., vol. 9, no. 8, pp. 979–986, 2015.
[55] V. Hinrichsen, G. Scholl, M. Schubert, and T. Ostertag, “Online Monitoring Of High Voltage Metal Oxide Arresters By
Wireless Passive Surface Acoustic Wave Temperature Sensors,” in High Voltage Engineering Symposium, 1999, pp. 238–
241.
[56] Y. Wen and C. Zhou, “A novel method for predicting the lifetime of MOV,” IEEE Trans. Power Deliv., vol. 19, no. 4, pp.
1688–1691, 2004.
[57] K. Kannus and K. Lahti, “Evaluation of the operational condition and reliability of surge arresters used on medium voltage
networks,” IEEE Trans. Power Deliv., vol. 20, no. 2 I, pp. 745–750, 2005.
[58] “SIOV metal oxide varistors,” EPCOS AG. 2008.
[59] J. Woodworth, “Arrester Condition Monitors,” ArresterWorks. 2012.
[60] M. A. Ponce et al., “Influence of degradation on the electrical conduction process in ZnO and SnO2-based varistors,” J.
Appl. Phys., vol. 108, no. 7, pp. 1–7, 2010.
[61] X. Yan, Y. Wen, and X. Yi, “Study on the resistive leakage current characteristic of MOV surge arresters,” in IEEE/PES
Transmission and Distribution Conference and Exhibition, 2002, vol. 2, pp. 683–687.
[62] D. Roethemeyer and D. Yankaskas, “Evolution of Motor and Variable Frequency Drive Technology,” Summer Study
Energy Effic. Ind. Conf. Proc., pp. 541–552, 1995.
[63] Global Efficiency Intelligence, “Infographic: The Profile of Energy Use in Industrial Motor Systems.” [Online]. Available:
https://www.globalefficiencyintel.com/new-blog/2017/infographic-energy-industrial-motor-systems.
[64] K. Marston, “Problems VFDs cause and cable types that help solve them,” 2015. [Online]. Available:
https://www.motioncontroltips.com/problems-vfds-cause-cable-types-help-solve/.
[65] A. Von Jouanne, P. Enjeti, and W. Gray, “Application issues for PWM adjustable speed AC motor drives,” IEEE Ind. Appl.
Mag., vol. 2, no. 5, pp. 10–18, 1996.
[66] A. Von Jouanne, P. Enjeti, and W. Gray, “The effect of long motor leads on PWM inverter fed AC motor drive systems,” in
Applied Power Electronics Conference and Exposition, 1995, vol. 2, no. 1, pp. 592–597.
[67] I. Tilea and C. Munteanu, “Overvoltage analysis in inverter driven induction motors,” in EPE 2012 - Proceedings of the
2012 International Conference and Exposition on Electrical and Power Engineering, 2012, no. Epe, pp. 509–512.
[68] J. A. De Castro, H. De Paula, and B. J. Cardoso Filho, “A new configuration for motor drive systems based on long cable
connection between rectifier and inverter - Part I: Analysis of the losses reduction and copper economy,” in 2009 Brazilian
 References 176

Power Electronics Conference, COBEP2009, 2009, pp. 848–855.

[69] H. Vang and M. Chiari, “An Improved Approach for Connecting VSD and Electric Motors,” Schneider Electric, 2013.
[70] A. Mbaye and T. Lebey, “Analytical approach of PD activity in low voltage motors fed by inverters [adjustable speed
drives],” pp. 161–164, 2002.
[71] T. G. Arora, M. M. Renge, and M. V. Aware, “Effects of switching frequency and motor speed on common mode voltage,
common mode current and shaft voltage in PWM inverter-fed induction motors,” Proc. 2017 12th IEEE Conf. Ind. Electron.
Appl. ICIEA 2017, vol. 2018-Febru, pp. 583–588, 2018.
[72] E. Sharifi-ghazvini, “Analysis of Electrical and Thermal Stresses in the Stress Relief System of Inverter Fed Medium
Voltage Induction Motors,” 2010.
[73] M. Mechlinski, S. Schroder, J. Shen, and R. W. De Doncker, “Grounding Concept and Common-Mode Filter Design
Methodology for Transformerless MV Drives to Prevent Bearing Current Issues,” IEEE Trans. Ind. Appl., vol. 53, no. 6, pp.
5393–5404, 2017.
[74] S. P M, D. B. Banakara, and S. Reddy, “Modeling, Simulation and Analysis of Bearing Voltage and Bearing Current in
PWM Multilevel Inverter Fed Induction Motor with Different HP Ratings,” Ijireeice, vol. 5, no. 4, pp. 92–99, 2017.
[75] H. De Paula, W. P. De Almeida, P. Pereira, A. V. Rocha, and B. J. C. Filho, “Driving AC motors through long distances
with DC transmission: Experimental results,” 2014 IEEE Ind. Appl. Soc. Annu. Meet. IAS 2014, pp. 1–7, 2014.
[76] R. M. Tallam, G. L. Skibinski, T. A. Shudarek, and R. A. Lukaszewski, “Integrated differential-mode and common-mode
filter to mitigate the effects of long motor leads on AC drives,” IEEE Trans. Ind. Appl., vol. 47, no. 5, pp. 2075–2083, 2011.
[77] S. Suwankawin, P. Pairodamonchai, S. Sangwongwanich, C. Sukhapap, and W. Tearwattanarattikal, “Destruction by charge
current and its solution for PWM inverter feeding multiple motors through long cables,” Conf. Rec. - IAS Annu. Meet. (IEEE
Ind. Appl. Soc., vol. 2, pp. 1382–1389, 2005.
[78] A. L. Julian, G. Oriti, and T. A. Lipo, “Elimination of common-mode voltage in three-phase sinusoidal power converters,”
IEEE Trans. Power Electron., vol. 14, no. 5, pp. 982–989, 1999.
[79] M. Kuschke and K. Strunz, “Transient cable overvoltage calculation and filter design: Application to onshore converter
station for hydrokinetic energy harvesting,” IEEE Trans. Power Deliv., vol. 28, no. 3, pp. 1322–1329, 2013.
[80] S. Amarir and K. Al-Haddad, “A new mathematical tool to investigate the influence of cable characteristics and IGBT fast
switching on voltage transients and differential mode currents for PWM drives,” Math. Comput. Simul., vol. 81, no. 2, pp.
180–193, 2010.
[81] A. Boglietti and E. Carpento, “Induction Motor High Frequency Model,” in IEEE Industry Applications Conference 1999,
1999, pp. 1551–1558.
[82] M. Moreau, N. Idir, and P. Le Moigne, “Modeling of Conducted EMI in Adjustable Speed Drives,” IEEE Trans.
Electromagn. Compat., vol. 51, no. 3, pp. 665–672, 2009.
[83] M. Schinkel, S. Weber, S. Guttowski, W. John, and H. Reichl, “Efficient HF Modeling and Model Parameterization of
Induction Machines for Time and Frequency Domain Simulations,” in Twenty-First Annual IEEE Applied Power
Electronics Conference and Exposition, 2006. APEC ’06., 2006, pp. 1181–1186.
[84] Y. Weens, N. Idir, J. J. Franchaud, and R. Bausiere, “High frequency model of a shielded 4-wire energy cable,” in 2005
European Conference on Power Electronics and Applications, 2005, vol. 2005, pp. 10 pp.-P.10.
[85] H. Akagi and I. Matsumura, “Overvoltage mitigation of inverter-driven motors with long cables of different lengths,” IEEE
Trans. Ind. Appl., vol. 47, no. 4, pp. 1741–1748, 2011.
[86] A. F. Moreira, T. A. Lipo, G. Venkataramanan, and S. Bernet, “High-frequency modeling for cable and induction motor
overvoltage studies in long cable drives,” IEEE Trans. Ind. Appl., vol. 38, no. 5, pp. 1297–1306, 2002.
[87] L. Wang, C. N. M. Ho, F. Canales, and J. Jatskevich, “High-frequency modeling of the long-cable-fed induction motor drive
system using TLM approach for predicting overvoltage transients,” IEEE Trans. Power Electron., vol. 25, no. 10, pp. 2653–
2664, 2010.
[88] A. F. Moreira, T. A. Lipo, G. Venkataramanan, and S. Bernet, “High frequency modeling for cable and induction motor
overvoltage studies in long cable drives,” in 2001 IEEE Industry Applications Conference. 36th IAS Annual Meeting (Cat.
No.01CH37248), 2001, vol. 3, pp. 1787–1794.
 References 177

[89] L. Giuntini, E. Blondel, and G. Mellini, “Surge protection for double-conversion UPS,” pp. E21–E21, 2012.
[90] Q. Zhou, F. Liu, Y. Cheng, and Y. Zhao, “Experimental study on the surge protection of converters in direct-driven
permanent-magnetic wind turbines,” 2015 Int. Symp. Light. Prot. XIII SIPDA 2015, pp. 186–190, 2015.
[91] R. W. Hotchkiss, “Surge Protection of Low Voltage AC Drives – Application Note,” in 2014 IEEE PES General Meeting
Conference & Exposition, 2014, pp. 3–5.
[92] K. Hinds and A. Saturno, “VFD Protection,” in Intelligent Ships Symposium, 2013.
[93] K. K. Yuen and H. S. Chung, “A Low-Loss ‘RL-Plus-C’ Filter for Overvoltage Suppression in Inverter-Fed Drive System
With Long Motor Cable,” IEEE Trans. Power Electron., vol. 30, no. 4, pp. 2167–2181, 2015.
[94] J. He, G. Y. Sizov, P. Zhang, and N. A. O. Demerdash, “A Review of Mitigation Methods for Overvoltage in Long-Cable-
Fed PWM AC Drives,” in 2011 IEEE Energy Conversion Congress and Exposition, 2011, pp. 2160–2166.
[95] A. F. Moreira, P. M. Santos, T. A. Lipo, and G. Venkataraman, “Filter networks for long cable drives and their influence on
motor voltage distribution and common-mode currents,” IEEE Trans. Ind. Electron., vol. 52, no. 2, pp. 515–522, 2005.
[96] J. K. Steinke, “Use of an LC filter to achieve a motor-friendly performance of the PWM voltage source inverter,” IEEE
Trans. Energy Convers., vol. 14, no. 3, pp. 649–654, 1999.
[97] Y. Sozer, D. A. Torrey, and S. Reva, “New inverter output filter topology for PWM motor drives,” IEEE Trans. Power
Electron., vol. 15, no. 6, pp. 1007–1017, 2000.
[98] S. J. Kim and S. K. Sul, “Novel filter design for suppression of high voltage gradient in voltage-fed PWM inverter,” Conf.
Proc. - IEEE Appl. Power Electron. Conf. Expo. - APEC, vol. 1, pp. 122–127, 1997.
[99] Y. Jiang, W. Wu, Y. He, H. S. H. Chung, and F. Blaabjerg, “New Passive Filter Design Method for Overvoltage
Suppression and Bearing Currents Mitigation in a Long Cable Based PWM Inverter-Fed Motor Drive System,” IEEE Trans.
Power Electron., vol. 32, no. 10, pp. 7882–7893, 2017.
[100] M. M. Swamy and M. A. Baumgardner, “New Normal Mode dv/dt Filter With a Built-In Resistor Failure Detection
Circuit,” IEEE Trans. Ind. Appl., vol. 53, no. 3, pp. 2149–2158, 2017.
[101] Y. Liu, L. Wang, H. Gao, H. Zhang, and D. Xu, “Overvoltage mitigation of submersible motors with long cables of different
lengths,” in 2014 17th International Conference on Electrical Machines and Systems (ICEMS), 2014, vol. 0, no. 1, pp. 638–
644.
[102] R. Mini, S. R. Resna, and M. N. Dinesh, “LC clamp filter for voltage reflection due to long cable in induction motor drives,”
in 2014 Annual International Conference on Emerging Research Areas: Magnetics, Machines and Drives, AICERA/iCMMD
2014 - Proceedings, 2014, pp. 1–6.
[103] A. Von Jouanne, D. A. Rendusara, P. N. Enjeti, and J. W. Gray, “Filtering techniques to minimize the effect of long motor
leads on pwm inverter-fed ac motor drive systems,” IEEE Trans. Ind. Appl., vol. 32, no. 4, pp. 919–926, 1996.
[104] M. Smochek, A. F. Pollice, M. Rastogi, and M. Harshman, “Long cable applications from a medium-voltage drives
perspective,” IEEE Trans. Ind. Appl., vol. 52, no. 1, pp. 645–652, 2016.
[105] K. K. F. Yuen and H. S. H. Chung, “Use of synchronous modulation to recover energy gained from matching long cable in
inverter-fed motor drives,” IEEE Trans. Power Electron., vol. 29, no. 2, pp. 883–893, 2014.
[106] J. Ström, J. Korhonen, J. Tyster, and P. Silventoinen, “Active du / dt — New Output-Filtering Approach for Inverter-Fed
Electric Drives,” IEEE Trans. Ind. Electron., vol. 58, no. 9, pp. 3840–3847, 2011.
[107] A. Bhakthavachala, S. T. Kalyani, and K. Anuradha, “A Simplified Filter Topology for Compensating Common Mode
Voltage and Electromagnetic Interference in Induction Motor Drives,” Energy Procedia, vol. 117, pp. 377–384, 2017.
[108] R. Ortega, A. de Rinaldis, G. Escobar, and M. W. Spong, “A hybrid active filter implementation of an overvoltage
suppression scheme,” in 2004 IEEE International Symposium on Industrial Electronics, 2004, pp. 1141–1146 vol. 2.
[109] J. A. Pomilio, C. R. De Souza, L. Matias, P. L. D. Peres, and I. S. Bonatti, “Driving AC motors through a long cable: The
inverter switching strategy,” IEEE Trans. Energy Convers., vol. 14, no. 4, pp. 1441–1446, 1999.
[110] H. Xiong, J. Zhang, and A. Von Jouanne, “Control of Variable Frequency Drive PWM to Mitigate Motor Overvoltage Due
to Double Pulsing in Reflected Wave Phenomenon,” 2018 IEEE Energy Convers. Congr. Expo. ECCE 2018, pp. 6563–
6570, 2018.
[111] S.-C. Lee, “Perfect wave oscillation suppression with reflection coefficients adaptation in PWM motor drives with long
 References 178

feeding cables,” IEE Proc. - Electr. Power Appl., vol. 153, no. 3, p. 467, 2006.
[112] J. A. De Castro, H. De Paula, and B. J. Cardoso Filho, “Avoiding undesirable high-frequency phenomena in long cable
drives: Rectifier-to-inverter connection through long DC cable - Part I: Evaluation of the losses reduction and copper
economy,” IECON Proc. (Industrial Electron. Conf., pp. 1045–1050, 2009.
[113] A. Rocks and V. Hinrichsen, “Overvoltage Protection of inverter-fed drives with the help of energy varistors – dimensioning
and lifetime considerations,” in 2008 Applied Power Electronics Conference and Exposition (APEC), 2008, pp. 860–865.
[114] A. Rocks and V. Hinrichsen, “Application of varistors for overvoltage protection of machine windings in inverter-fed
drives,” in 2007 IEEE International Symposium on Diagnostics for Electric Machines, Power Electronics and Drives,
SDEMPED, 2007, pp. 335–340.
[115] A. Rocks and V. Hinrichsen, “Effect of varistors for overvoltage protection in inverter-fed drives on the voltage distribution
in the stator winding,” Conf. Rec. IEEE Int. Symp. Electr. Insul., pp. 624–627, 2008.
[116] A. Rocks and V. Hinrichsen, “Overvoltage protection of inverter-fed drives with the help of energy varistors: Dimensioning
rules for consideration of different cable types,” Conf. Proc. - IEEE Appl. Power Electron. Conf. Expo. - APEC, pp. 330–
335, 2009.
[117] R. W. Hotchkiss, “Modes of Protection within Electrical Systems for Application of Surge Suppression by Ronald W.
Hotchkiss VP/Engineering, Surge Suppression Incorporated,” 2005.
[118] V. S. Brito, G. R. S. Lira, E. G. Costa, and M. J. A. Maia, “A Wide-Range Model for Metal-Oxide Surge Arrester,” IEEE
Trans. Power Deliv., vol. 33, no. 1, pp. 102–109, 2018.
[119] C. Widemann, S. Scheier, S. Frei, and W. Mathis, “Modeling the static and dynamic behavior of multi-layer varistors in the
threshold voltage region depending on the DC operating point,” 2014 21st IEEE Int. Conf. Electron. Circuits Syst. ICECS
2014, pp. 698–701, 2015.
[120] S. Scheier, D. Deelmann, S. Frei, C. Widemann, and W. Mathis, “A combined time and frequency domain characterization
method for modeling of overvoltage protection elements,” IEEE Int. Symp. Electromagn. Compat., vol. 2015-Septm, pp.
1351–1356, 2015.
[121] L. Muremi and P. Bokoro, “Assessing the correlation between impedance and reference voltage of varistor arresters using
linear regression model,” Proc. IECON 2018 - 44th Annu. Conf. IEEE Ind. Electron. Soc., vol. 1, pp. 109–113, 2018.
[122] J. T. S. J. Bak-Jensen, B. Bak-Jensen, S. Damsgaard Mikkelsen, “EXAMINATIONS OF ZnO-VARISTORS IN THE
LFAKAGE REGION WITH REG- TO HIGH FRFQUENCY STRESSES,” no. 2, pp. 191–196, 1993.
[123] E. Shulzhenko, M. Rock, M. Birle, and C. Leu, “Applying of surge arresters in power electronic network components,” in
2014 International Conference on Lightning Protection (ICLP), 2014, vol. 103, no. 3, pp. 904–910.
[124] “Gas Discharge Tube (GDT) Products.” Littlefuse, 2017.
[125] “2027 Series - 2-Pole Gas Discharge Tube 2027 Series.” Bourns, pp. 1–2, 2020.
[126] M. Pejovic, M. Pejovic, R. Maric, L. Timotijevic, and K. Stankovic, “Delay response of gas discharge tubes,” Prog.
Electromagn. Res. Symp., no. November, pp. 1156–1159, 2012.
[127] T. Ardley, “First Principles of a Gas Discharge Tube ( GDT ) Primary Protector Table of Contents,” Bourns. 2008.
[128] J. He, C. Li, J. Hu, and R. Zeng, “Deep Suppression of Switching Overvoltages in AC UHV Systems Using Low Residual
Arresters,” IEEE Trans. Power Deliv., vol. 26, no. 4, pp. 2718–2725, Oct. 2011.
[129] P. S. S. Holenarsipur, D. J. Christofersen, and N. Mohan, “Thyristor-switched arrester for improved protection of power
system equipment,” IEEE Eng. Soc. Winter Meet., vol. 2, pp. 1009–1014, 1999.
[130] “IEEE Guide for the Application of Thyristor Surge Protective Device Components IEEE Power and Energy Society,” no.
April. 2013.
[131] “Trisil TM for telecom equipment protection.” STMicroelectronics, 2012.
[132] C. Blake and C. Bull, “IGBT or MOSFET: Choose wisely,” Infineon Technologies. 2018.
[133] “Surge currents for IGBT diodes,” ABB. 2013.
[134] F. Schork, R. Brocke, and M. Rock, “Surge current capability of power electronics for surge protection,” in PCIM Asia 2016
- International Exhibition and Conference for Power Electronics, Intelligent Motion, Renewable Energy and Energy
Management, 2016, no. June, pp. 28–30.
 References 179

[135] T. Basler, J. Lutz, and R. Jakob, “IGBTs conducting diode-like surge currents,” Proc. Int. Symp. Power Semicond. Devices
ICs, no. 2, pp. 103–106, 2014.
[136] P. Hofstetter and M. Bakran, “Comparison of the Surge Current Ruggedness between the Body Diode of SiC MOSFETs and
Si Diodes for IGBT Surge current measurements with SiC MOSFET body diodes,” in CIPS 2018 – 10th International
Conference on Integrated Power Electronics Systems Comparison, 2018, pp. 355–361.
[137] L. Knoll et al., “Robust 3.3kV silicon carbide MOSFETs with surge and short circuit capability,” in 2017 29th International
Symposium on Power Semiconductor Devices and IC’s (ISPSD), 2017, pp. 243–246.
[138] S. Yin, Y. Gu, S. Deng, X. Xin, and G. Dai, “Comparative Investigation of Surge Current Capabilities of Si IGBT and SiC
MOSFET for Pulsed Power Application,” IEEE Trans. Plasma Sci., vol. 46, no. 8, pp. 2979–2984, 2018.
[139] K. Askan and E. Industries, “Surge Current Capability of IGBTs Used in Low Voltage DC / AC Hybrid Circuit Breaker
Selection of Device Under Test ( DUT ) Determination of Output Charac- teristics and Non-destructive SOA,” in CIPS
2018; 10th International Conference on Integrated Power Electronics Systems, 2018, pp. 192–197.