P OLITECNICO DI M ILANO

D IPARTIMENTO DI E LETTRONICA , I NFORMAZIONE E B IOINGEGNERIA

D OCTORAL P ROGRAMME I N E LECTRICAL E NGINEERING

E LECTRIC F IELD M ODELLING IN

HVDC I NSULATION S YSTEMS

Doctoral Dissertation of:

Pasquale Cambareri

Supervisor:
Prof. Luca Di Rienzo
Co-supervisor:
Prof. Carlo de Falco
The Chair of the Doctoral Programme:
Prof. Marco Mussetta

Year 2022 – Cycle XXXV

 I would like to thank Prof. Luca Di Rienzo (DEIB, Politecnico di Milano), Prof.
Carlo de Falco (MOX, DMAT, Politecnico di Milano), Dr. Paolo Seri (LIMES, DEI,
Università di Bologna) and Prof. Gian Carlo Montanari (CAPS, Florida State Univer-
sity) for their support and guidance during these three years.
 Abstract

T HE landscape of electrical infrastructures has been evolving for quite a long time.
Climate change demands for an extensive and responsible exploitation of renew-
able energy sources to generate electric power, whose characteristics differ sub-
stantially from the traditional fossil-fuels-based power generation. Moreover, the need
for a more flexible electrical infrastructure is pushing towards an increasing number of
interconnections among systems that differ substantially in size and nature. These are
only few of the factors that led, over the years, to a gradually renewed interest in DC
distribution and transmission systems, especially in the high voltage range.
Modern DC, and particularly HVDC, systems are immersed in a harsh environment.
The once AC-dominant infrastructure no longer exists, and components such as cables,
joints and motors are subject to higher-than-ever stresses. This situation is particularly
challenging for the insulation of such components, which is notoriously the weakest
constituent. One of the great challenges in insulation design is the estimation of the
expected life duration under normal and extraordinary operating conditions. A wrong
estimate may cause a dielectric breakdown much earlier than expected, undermining
the integrity of the electrical infrastructure.
One of the key factors that determines the life duration of an insulation system is
the electric field, which is the prime mover behind a series of ageing processes that
cause the fast degradation of the insulation system over time. An accurate calculation
of the electric field, both in transient and steady state conditions, is possible only with
the joint help of experimental measurements that unveil the physical properties of the
insulation materials, and of mathematical models that allow to investigate alternative
scenarios beyond the experimental ones through numerical simulations.
The present work has the goal of investigating the properties of two classes of insu-
lating materials: oil-paper and polymeric insulation, which are widely used in HVDC
systems.
Oil-paper insulation has been investigated by means of the Occhini equivalent cir-
cuit (which is the circuital representation of the well-known Debye extended model),
according to which the dynamic is dominated by dielectric polarization processes. It
has been confirmed that oil-paper dynamic is dominated by such processes. A numer-

I
ical scheme based on finite differences has been developed to study the field transient
in an HVDC cable. The results have shown that the polarization processes slow down
the field transient and that, in particular, this is beneficial for the cable, because the
maximum field is reached only at DC steady state.
Polymeric insulation modelling requires a different physical basis, because the Oc-
chini circuit is not able to model this type of insulation. A model based on charge
trapping and detrapping processes, under the framework of the band theory of solids,
has been derived to describe the transient behaviour observed in thin specimens of
polypropylene and cross-linked polyethylene. The energy band structure separates deep
and shallow traps. The deep traps have been modelled with a continuous density of
states, while the contribution of the shallow traps to the transient current has been mod-
elled empirically with a power law. In the model, the capture and release of charge
carriers is attributed to the deep traps, while the shallow traps impact on the mobility
(thus, also on the conductivity) of the material. The two types of traps have been as-
sumed to be non-interacting, thus two independent sub-models could be adopted. The
model was shown to provide a good fitting of the experimental curves.
In the last part of this work the topic of partial discharges modelling has been cov-
ered, which constitute one of the factors that cause accelerated insulation ageing. The
typical modelling approach developed by Niemeyer works very well with uniform elec-
tric field profiles, but its application to non-uniform profiles is cumbersome. An alter-
native approach, called the ks approach, is proposed here, which is simpler to apply in
the case of non-uniform field profiles and it is more general. It is shown that the pro-
posed approach yields a correct evaluation of the partial discharge inception voltage,
thus it is a viable alternative to Niemeyer’s approach.
The work closes with an investigation of the electric field transient in the case of an
internal cylindrical cavity embedded within a polymeric sample. The transient is eval-
uated using the polymeric insulation model previously developed and the traditional
model based on the RC equivalent circuit. The study highlights that the polymeric in-
sulation model predicts a substantially different dynamic with respect to the RC model,
especially in terms of the transient duration.

II
 List of Abbreviations and Symbols

f˙(t) and ġ Time derivative of a function with respect to time

E, D, P and J Electric, displacement, polarization and current density vector fields
E, D, P and J Corresponding magnitudes of the previous fields
H Heaviside’s step function
δ Dirac’s delta function
ω Angular frequency
F(ω) Fourier transform of function F (t)
ε0 Dielectric permittivity of vacuum
ε∞ High-frequency dielectric permittivity
IPM Inverse Power Model
HV High Voltage
GND Ground
PDC Polarization and Depolarization Currents
PD Partial Discharges
PP Polypropylene
XLPE Cross-Linked Polyethylene
PDIV Partial Discharge Inception Voltage
FD Finite Differences
SC Space Charge

III
 Contents

1 The Role of Electric Fields in DC Insulation Systems 1

1.1 Intrinsic Ageing: Phenomenological Life Models . . . . . . . . . . . . 1
1.2 Extrinsic Ageing Factors and Accelerated Ageing . . . . . . . . . . . . 3
1.3 Electric Field Modelling as a Tool to Support The Study of Insulation
Ageing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Polarization Processes and Electric Field Transient 7

2.1 The Classical Theory of Dielectrics: an Overview . . . . . . . . . . . . 7
2.2 The Occhini Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Fitting the Occhini Model Parameters from PDC Measurements . . . . 14
2.4 Equivalency between the Occhini Model and the Finite Difference Dis-
cretization of the Cable Geometry . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Finite Difference Scheme for Semi-Discretization in Space . . . 21
2.4.2 Equivalence between the Semi-Discretized Model and a Sequence
of Occhini Circuits . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Impact of Polarization Processes on the Electric Field Transient in Paper-
Insulated DC Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Modelling Issues in Polymers 33

3.1 Non-Linear Response . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Inadequacy the Occhini Model for Polymers . . . . . . . . . . . . . . . 35
3.2.1 First Attempt: Fitting the Depolarization Current with 3 R-C
Branches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2.2 Second Attempt: Fitting the Polarization Current with 3 R-C
Branches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.3 Third Attempt: Fitting the Depolarization Current with more than
3 R-C Branches . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Trapping and Detrapping in Polymers: a Circuital Model 41

4.1 Preliminary Considerations on the Model . . . . . . . . . . . . . . . . 41
4.2 Band Theory of Solids: Key Concepts . . . . . . . . . . . . . . . . . . 42

V
Contents

4.2.1 Donors and Acceptors . . . . . . . . . . . . . . . . . . . . . . . 44

4.2.2 Trapping Levels . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2.3 Surface Trapping Levels . . . . . . . . . . . . . . . . . . . . . . 46
4.2.4 Metallic Contacts . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2.5 Application of an External Electric Field . . . . . . . . . . . . . 47
4.2.6 Injection and Extraction . . . . . . . . . . . . . . . . . . . . . . 48
4.2.7 Drift and Diffusion . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 Fundamental Assumptions, Modelling of Deep and Shallow Traps . . . 50
4.3.1 Deep Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3.2 Shallow Traps . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.4 The Journey of an Electron . . . . . . . . . . . . . . . . . . . . . . . . 54
4.5 The Equivalent Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.6 Experimental Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.7 Density of States of the Deep Traps . . . . . . . . . . . . . . . . . . . 64
4.8 Fitting of PDC Measurements . . . . . . . . . . . . . . . . . . . . . . 66
4.8.1 PP1 and PP2 Transient Reconstruction . . . . . . . . . . . . . . 67
4.8.2 PP3 Transient Reconstruction . . . . . . . . . . . . . . . . . . . 69
4.8.3 XLPE Transient Reconstruction . . . . . . . . . . . . . . . . . . 70
4.9 Concluding Comments . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5 Partial Discharges Modelling: the ks Approach 75
5.1 The Origin of a Discharge . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Niemeyer’s Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3 The ks Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4 Surface Partial Discharge . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4.1 Simulation Results and PDIV Calculation . . . . . . . . . . . . 85
5.4.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.5 Internal Partial Discharge . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.5.1 PDIV Calculation with the Traditional RC Model . . . . . . . . 93
5.5.2 PDIV Calculation Using the Polymer Model . . . . . . . . . . . 97
5.5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

6 Conclusions 103
Bibliography 105

VI
 CHAPTER 1
The Role of Electric Fields in DC Insulation
Systems

This chapter lays down the motivations that led to the research presented
in this work. The first two sections recall the electric field role in insula-
tion ageing. The third section illustrates how improved electric field models
can support the study of insulation ageing. The content of this introductory
chapter is derived from [1].

1.1 Intrinsic Ageing: Phenomenological Life Models

The weakest constituent of an electric system is the electrical insulation. At the very
moment in which a dielectric breakdown event bursts into the insulation, the encom-
passing infrastructure activates a chain of procedures aimed at isolating the faulty com-
ponent and minimizing the impact on the entire system. Sometimes the fault is tem-
porary and gets fixed in relatively short time, sometimes the damage is more serious.
Nevertheless, the entire infrastructure is weakened. The electrical insulation, then, is
strictly entangled to the electrical system’s fragility, intended as the risk of going out
of service after sudden events such as a dielectric breakdown.
When thinking about an insulation system, the breakdown event is the point of no re-
turn. Like a ship sailing at night, the dielectric breakdown is the lighthouse marking the
borderline between the sea and the coastline. An erroneous evaluation of the distance,
and the ship runs aground. A class of the tools developed by researchers and engineers
to “measure the distance from the lighthouse” are the life models. The underlying idea
is simple: the insulation is exposed, over its life, to a number of different stresses, such
as thermal, mechanical and electrical ones. As a consequence, the insulation suffers
a degradation of its chemical and physical properties, called intrinsic ageing, that will

1
Chapter 1. The Role of Electric Fields in DC Insulation Systems

log V
V0
n1
n2
Vr
t0 t1 t2 log t
Figure 1.1: Interpretation of the voltage endurance coefficient, n. In this plot n1 > n2 . Vr is the rated
voltage of the insulation system.

cause, at a certain time in the future, its failure due a dielectric breakdown. The aim of
a life model is to estimate the time to failure for a given applied stress.
One of the simplest life models is the inverse power model (IPM), that links the time
to failure, t, to the applied voltage, V :
 −n
V
t = t0 , (1.1)
V0
where t0 and V0 are some reference time to failure and applied voltage. The model
clearly states (n > 0) that the higher the applied voltage, the shorter the time to failure.
The n at the exponent of the above equation is the voltage endurance coefficient. To
understand its meaning, let us refer to Figure 1.1. The IPM relationship is usually
shown on a log V versus log t plot, where the relation between t and V is linear:
1

log V = log V0 − log t − log t0 . (1.2)
n
The reference voltage V0 is typically much higher that the rated voltage of the insulation
system, so that the reference time to failure, t0 , is relatively short and the IPM line starts
on the top-left region of the plot, Figure 1.1. This decision is a consequence of the fact
that the IPM is a phenomenological model, thus the parameters of equation (1.1) are
fitted from experimental tests run on the insulation system. Since the time to failure
of some systems can be very long, for example 30 or 40 years in the case of cables,
the IPM parameters are retrieved from accelerated life tests, with an applied voltage
higher than the rated one that causes the insulation to fail sooner. The time to failure
at rated voltage then is extrapolated from the fitted model. Figure 1.1 shows two IPM
lines with different voltage endurance coefficients, n1 > n2 . The line with the greater
coefficient, n1 , has a longer time to failure at rated voltage, Vr , than the line with the
lower coefficient, n2 . This is the meaning of n.
The applied voltage, V , in (1.1) is sometimes replaced with the insulation electric
field, E, hence the IPM is written as
 −n
E
t = t0 . (1.3)
E0
The electric field can be the either the average or the maximum one. It depends on the
applied voltage and on the geometry of the insulation system. The voltage endurance
coefficient in (1.3) may, in general, be different from the one in (1.1).

2
 1.2. Extrinsic Ageing Factors and Accelerated Ageing

The IPM is a simple and practical tool for the estimation of the expected lifetime of
an insulation system. However, one should consider that the time to failure is, in fact, a
random variable: if the same voltage is applied to two different prototypes of the same
system, it is highly unlikely that they will both fail at the same time. The best statistical
description of the time to failure is the Weibull distribution. The cumulative distribution
function is "   #
β
t
F (t) = 1 − exp − , (1.4)
α
where α is the scale parameter and β is the shape parameter. Time t = α coincides
with the time to failure at 63% probability. The functional form of (1.4) is, however,
incomplete, as there is no indication of the influence of the applied stresses on the
failure probability. A typical empirical model for the scale parameter accounting for
electrothermal stresses is
"  #  −n
1 1 E
α(T, E) = α0 · exp −a − · , (1.5)
T T0 E0

where T is the temperature and E is the applied average electric field. The first factor
in (1.5), α0 , is the scale parameter at reference temperature and field, T0 and E0 . The
exponential factor is an Arrhenius law, whose employment in the context of insula-
tion ageing was originally proposed by Dakin in [2]. Dakin’s theory treats insulation
ageing as a thermally activated chemical reaction, which justifies the employment of
Arrhenius’ law. It should be mentioned that while this law works well in practice, the
physical interpretation of its parameters is turbid. A more refined model that uses the
reaction rate theory of Eyring [3] was discussed by Crine in [4]. The third factor in
(1.5) has the same functional form of the IPM (1.3), and n is still the voltage endurance
coefficient. As for the shape parameter, β, no significant sensitivity to the applied stress
has been observed over time, thus it can be considered constant.

1.2 Extrinsic Ageing Factors and Accelerated Ageing

The models presented in the previous section assume that the insulation ages uniformly
throughout its volume. There are, however, two particularly relevant phenomena that
differ substantially under this aspect, since they are confined to a narrow volume of
the insulation. Such localized phenomena are the space charge (SC) accumulation and
partial discharges (PD). Both phenomena are characterized by a high energy density
that, with different mechanisms, can break the intermolecular and intramolecular bonds
of the insulating material, creating defects that grow over time and become the precur-
sors of the electrical treeing that eventually brings the insulation to breakdown [5–7].
Space charge and partial discharges fall under the class of extrinsic ageing. The lo-
calized damage caused by these ageing factors has the unfortunate consequence of ac-
celerating the ageing rate of the insulation, causing a premature failure of the entire
system. The matter will now be illustrated qualitatively, with the help of the IMP (1.3)
and Figure 1.2.
Let us consider an insulation system with a design average electric field Ed (which
depends on the rated voltage, Vr ) and an expected time to failure td at E = Ed . The

3
Chapter 1. The Role of Electric Fields in DC Insulation Systems

log E
E0
B
Ci

nPD n
Ed
C A

t0 tSC ti tPD td log t

Figure 1.2: Qualitative illustration of the accelerated ageing caused by extrinsic ageing phenomena,
namely space charge and partial discharges. Point A is the failure event expected by the design.
Point B is the failure event with space charge. Point Ci is the PD triggering event, after which the
IPM diagram has a bifurcation: the original line with slope n, that assumes the absence of PD
activity, and the line with slope nP D < n, that accounts for the PD activity. Point C is the failure
event at design field expected under PD activity.

failure event expected by design in Figure 1.2 is pinpointed by point A, with coordinates
(td , Ed ). Let us now consider the implication of space charge and partial discharges on
the insulation failure.

Space charge Dakin’s idea was that insulation ageing is a thermally activated chem-
ical reaction with a certain activation energy. Space charges cluster into local-
ized sites of the insulation volume, forming small regions of high charge density.
The accumulated charge provides an additional contribution of potential energy,
in electrostatic and electromechanical form, that modifies the ground state of the
localized site and, in particular, lowers the activation energy of the degradation
processes that cause ageing [8]. In other words, such processes are still activated
thermally, but the presence of space charge ease their activation, hence the insu-
lation ages faster, causing the failure event at a time tSC which is shorter then the
one expected by design, Figure 1.2.

Partial discharges Partial discharges occur in cavities embedded in the insulation vol-
ume at the nano-metric or micro-metric scale [9, 10]. They can also occur on the
surface of the insulation system, especially at triple points between the insula-
tion, a conductive object and air [10, 11]. The partial discharge activity is initiated
when the electric field overcomes a threshold value called inception field. The
passing of this threshold may be either a transient or a steady-state condition. In
DC systems, partial discharges have been observed in both situations [12]. If the
partial discharges are incepted at time ti (point Ci in Figure 1.2), the accelerated
ageing caused by the activity is reflected in a reduction of the voltage endurance
coefficient. Therefore, the IPM curve has a bifurcation at point Ci , where a second
line with slope nP D stems from the original one, whose slope is n > nP D [13].
The line with slope nP D projects a failure event at design field, Ed , in point C of
Figure 1.2. The time to failure under PD activity, tP D , is shorter than the original
time to failure expected by design, td .

4
 1.3. Electric Field Modelling as a Tool to Support The Study of Insulation Ageing

1.3 Electric Field Modelling as a Tool to Support The Study of Insulation

Ageing
This brief introductory chapter has hopefully conveyed, in a rather simplified manner,
the idea that electric fields play a major role in the mechanisms that rule insulation
ageing and ultimately lead to breakdown. Phenomenological life models for intrinsic
ageing are derived under the assumption of uniform degradation within the insulation
volume. Models like the IPM adopt the average insulation field as the starting quantity
to estimate the time to failure. On the other hand, extrinsic ageing factors, namely
space charge and partial discharges, rely on significantly different quantities. Space
charge causes localized field magnifications, hence the maximum electric field at such
“hotspots” is relevant. Partial discharges instead are driven by the local field at defects
or triple points, that can initiate a discharge if it overcomes a certain threshold. Since
both intrinsic and extrinsic ageing phenomena depend on the electric field intensity, the
ability to assess their impact on the insulation life is tied to the knowledge that one has
on electric field behaviour during transients and at steady state.
The development of physical and mathematical models to study the electric field
transient in DC insulation systems is the main topic of the present work. Chapter 2
discusses the electric field dynamic in oil-paper insulation, which is dominated by slow
polarization processes, namely dipolar and interfacial polarization [14]. A numerical
model based on the finite difference scheme has been developed to study the electric
field transient in HVDC cables with oil-paper insulation [15, 16]. Chapters 3 and 4
present a work that led to the development of a model to describe the dynamics of
polymeric insulators under the assumption of uniform electric field [17,18]. The model
assumes charge trapping and detrapping as the dominant processes, and it is derived
from the works of Schokly and Read [19], and Simmons and Tam [20–22]. Eventually,
chapter 5 develops an approach for partial discharges modelling that is alternative to the
well-known Niemeyer approach [10], and has the advantage of being simpler and more
general [11]. The chapter discusses a final example in which the model for polymeric
insulation is applied to the case study of internal partial discharges.
The problem of space charge modelling is not covered in this work.

5
 CHAPTER 2
Polarization Processes and Electric Field Transient

This chapter begins with a brief overview of the classical theory of dielec-
tric materials, that is needed to introduce the concepts of polarization pro-
cess and time-dependent dielectric permittivity. The Occhini model and a
technique to fit its parameters from polarization and depolarization current
measurements are presented as a tool to identify the dominant polarization
processes taking place in a dielectric material. The equivalence between the
Occhini model and the finite difference discretization of the geometry of a DC
cable provides the logical bridge to study the transient of the electric field in
the cable itself. Finally, the developed tools are used to describe the impact
of polarization processes on the electric field transient in paper-insulated
DC cables.

2.1 The Classical Theory of Dielectrics: an Overview

This section provides a brief overview of the key concepts developed in the classical
description of dielectric materials under DC excitation. This section has been extracted
from Jonscher’s book [14].
The study of the properties of dielectrics begins with the experimental observation
of the response of a certain material to the application of an external electric field. The
typical experimental setup, shown in Figure 2.1, involves a flat sample of the material
under investigation, with a certain area A and thickness d, placed between two metallic
electrodes. One of the electrodes is connected to a voltage source, the other one is
connected to the ground. The application of a voltage causes the flow of an electric
current between the two electrodes and through the bulk of the material. The current
is measured by an ammeter placed on the ground electrode. The goal of the theory,
then, is twofold. On one hand it has to provide a mathematical model that describes the

7
Chapter 2. Polarization Processes and Electric Field Transient

Electrode
AC or DC + Sample
voltage source -
Electrode

A Ammeter

Figure 2.1: Schematic of the typical experimental setup used to investigate the dielectric properties of a
material.

relation between the applied voltage and the measured current. On the other hand, it
has to attempt an explanation for the observed behaviour of the current, that is framed
in the context of quasi-static electric fields and on the dynamic of atoms, molecules and
free electric charges involved at a microscopic level.
The application of a voltage has another well-known consequence, that is the build
up of an electric charge at the interface between the electrodes and the dielectric. The
magnitude of this charge can be written in terms of the electric displacement field, D(t),
q(t) = AD(t). (2.1)
In (2.1) it is assumed that all the involved fields are uniform and have the same direction,
which is perpendicular to the surface of the electrodes and of the material. Therefore,
the displacement field is a scalar function of time. This is a very reasonable assumption
for the experimental setup described above. If charge q(t) is a consequence of the
applied voltage, the displacement field is a consequence of the applied electric field,
E(t). More precisely,
D(t) = ε0 E(t) + P (t), (2.2)
where ε0 is the dielectric permittivity of free space. In (2.2) the term ε0 E(t) describes
the instantaneous response of free space, while P (t), which is called polarization, de-
scribes the response of the material. P (t) is still related to the electric field, and in
general it represents the delayed response of the material. To clarify the difference
between instantaneous and delayed response, suppose that the electric field is
E(t) = E0 H(t), (2.3)
where E0 is the field amplitude and H(t) is the Heaviside step function. For t > 0, the
displacement field in (2.2) can be written as
Z t
D(t) = ε0 E0 + ε0 E0 f (τ ) dτ. (2.4)
0

In (2.4) f (τ ) is a dimensionless function called dielectric response function. A com-

parison with (2.1) implies that the first addend in (2.4) models a charge that arises at
the interface between the electrodes and the dielectric as soon as the field is applied
(instantaneous response). The second addend in (2.4) describes a cumulative process
that builds up the charge over time (delayed response). The dielectric response function
is the response of the material to the application of a step field and in order to provide
a consistent mathematical description it has to satisfy a few constraints:

8
 2.1. The Classical Theory of Dielectrics: an Overview

• f (t) = 0 for t < 0, as there should be no response of the material before the
application of a field (principle of causality),
R +∞
• f must be such that the 0 f (τ ) dτ is finite.
The expression of the second addend in (2.4) stems from the general definition of the
polarization as the convolution between the dielectric response function and the electric
field, [14], Z +∞
P (t) = ε0 f (τ )E(t − τ ) dτ. (2.5)
−∞

It is important to remark that (2.5) holds under the assumption that the polarization in
the dielectric material has a linear response.
The description of the material response in terms of the displacement field and of
the polarization has to be reconciled with the fact that the experiments allow to measure
a current, and not the charge q(t) in (2.1). The link between the two aspects is given by
the definition of electric current,
dq
ip (t) = . (2.6)
dt
The current that appears in (2.6) is called polarization current, since it is related to
the polarization P (t) through (2.1) and (2.2). Equation (2.6) highlights the fact that
the investigation of the properties of dielectric materials goes through a careful study
of the transient behaviour of ip (t). This current, however, is the outcome of several
processes occurring in parallel within the dielectric, that may also influence each other
in a non-negligible manner. The inference of the dielectric properties of a material then
is not an easy task, and it must be carried on with great care. For instance, alongside the
polarization processes that give rise to ip (t), dielectric materials exhibit also conduction
processes. The current density associated to such processes is

Jc (t) = σE(t), (2.7)

and the associated conduction current is

ic (t) = AJc (t) = AσE(t). (2.8)

Therefore, the total current measured in the experiments on dielectric materials is the
sum of conduction and polarization processes, that are assumed to be independent from
each other, at least as a first approximation,

im (t) = ic (t) + ip (t). (2.9)

Equation (2.9), along with the previous discussion, summarizes the descriptive task of
the theory, that relies on a series of generic time-dependent functions. The second task,
as already outlined, is to provide an explanation of the involved physical phenomena,
that allows to specify with greater detail the characteristics of the above functions.
The conduction current in dielectrics arises from an average motion of free charges
through the bulk in the direction of the applied electric field. Free charges in dielectrics
are made available by thermal excitation that separates, for example, the outer electrons

9
Chapter 2. Polarization Processes and Electric Field Transient

in the valence band from their atoms. The resulting effect is a temperature-dependent
electric conductivity for dielectrics, that is modelled by an Arrhenius law [14, 23–25],
 
W
σ(T ) = σ0 exp − , (2.10)
kB T
where σ0 is the conductivity at infinite temperature, W is the activation energy, T is the
temperature and kB is the Boltzmann constant.
The rise of the polarization current is ascribed to a finite displacement of charges in
the bulk of the material under the application of an electric field. There are several ways
in which the displacement can take place, and they are associated to different types of
polarizations:
• displacement of the negatively-charged electron cloud in atoms with respect to the
positive nucleus (electronic polarization),
• displacement of the positive and negative atoms within a molecule (atomic polar-
ization),
• alignment of polar molecules with the applied electric field (dipolar polarization),
• clustering of charges at the interfaces inside the material, such as the interface
between crystalline and amorphous regions in polymers, or between foils and oil
in impregnated paper insulation (interfacial polarization).
The properties of the polarization current are typically unveiled with an analysis of
the response of the material in the frequency domain. The most appropriate mathemat-
ical tool to adopt in this context is the Fourier transform. Therefore, considering the
expression of the displacement field, (2.2), and the general definition of the polariza-
tion, (2.5), in the time domain, it follows that

D(ω) = ε0 E(ω) + P(ω)

= ε0 E(ω) + ε0 χ(ω)E(ω)
 
= ε0 1 + χ(ω) E(ω) (2.11)

where D(ω), E(ω) and P(ω) are the Fourier transforms of the displacement, electric
and polarization fields, while the dielectric susceptibility χ(ω) is the Fourier transform
of the dielectric response function. The susceptibility is the quantity that retains the
properties of the material’s response and it is possible to define the dielectric permittiv-
ity as  
ε(ω) = ε0 1 + χ(ω) . (2.12)
The mathematical description of the properties of the dielectric, encoded by χ(ω)
and ε(ω), should, in principle, take into account the dynamics of atoms, molecules
and charges at a microscopic level. However, for the purpose of this thesis it is not
necessary to reach such a fine level of detail. It is sufficient to describe these processes
as a simple Debye relaxation mechanism [26], that can be modelled in the time domain
as, [14],
dP
τ + P = ε0 χE(t). (2.13)
dt
10
 2.1. The Classical Theory of Dielectrics: an Overview

In (2.13) τ and χ are the characteristic time constant and the (constant) susceptibil-
ity of the process, respectively. For electronic and atomic polarizations τ ≤ 10−12 s.
For dipolar polarization τ lies between 10−10 s and 101 s. For interfacial polarization τ
ranges between 102 s and 106 s. The values of τ and χ can be fitted from experimental
measurements, as it will be described in section 2.3. In the context of dielectric ma-
terials employed in DC insulation systems the relevant time scales are in the order of
minutes, hours and days. Therefore, electronic, atomic and possibly dipolar polariza-
tions can be considered by all means instantaneous, thus their associated polarization
takes the simpler form
P = ε0 χE(t). (2.14)
In this thesis a dielectric material will be described by a polarization with the following
components:
• the fast polarization due to electronic, atomic and the dipolar processes with a
time constant lower than 1 s,
P∞ (t) = ε0 χ∞ E(t), (2.15)
where χ∞ is the associated susceptibility;
• the slow polarization due to dipolar and interfacial processes with a time con-
stant higher than 1 s, which is modelled by a certain number of Debye relaxation
processes,
dPk
τk + Pk = ε0 χk E(t), (2.16)
dt
with k = 1, . . . , M .
The total polarization is, then, given by
M
X
P (t) = P∞ (t) + Pk (t). (2.17)
k=1

Equation (2.17), together with (2.15) and (2.16), is usually referred to as the extended
Debye model. It can be written in the frequency domain by applying again the Fourier
transform. To this aim, the transform of (2.15) is P∞ = ε0 χ∞ E(ω), while (2.16) yields
ε0 χ k
Pk (ω) = E(ω). (2.18)
1 + iωτk
Therefore, (2.17) becomes
 
M
X χk 
P(ω) = ε0 χ∞ + E(ω) = ε0 χ(ω)E(ω). (2.19)
k=1
1 + iωτk

The term between square brackets is the dielectric susceptibility, therefore from (2.11)
it follows that
 
M
X χk 
D(ω) = ε0 1 + χ∞ + E(ω) = ε(ω)E(ω). (2.20)
k=1
1 + iωτ k

11
Chapter 2. Polarization Processes and Electric Field Transient

The dielectric permittivity is given by the multiplication between ε0 and the term be-
tween square brackets in (2.20) and it can be written as
M
X ε0 χk
ε(ω) = ε∞ + . (2.21)
k=1
1 + iωτk

In (2.21) ε∞ = ε0 (1+χ∞ ) is the permittivity related to the instantaneous response of the

vacuum and of the fast polarization. The infinity symbol at the subscript of ε∞ indicates
that the properties of the fast polarization processes are found at very high frequencies
of the spectrum. If the step electric field (2.3) is applied, the inverse transform of (2.20)
yields, for t > 0,
D(t) = ε(t)E0 , (2.22)
where
M
X  
ε(t) = ε∞ + ε0 χk 1 − e−t/τk (2.23)
k=1

is the time-dependent dielectric permittivity.

2.2 The Occhini Model

Occhini and Maschio proposed in [23] a circuital model that is suitable to describe the
electric field transient in HVDC cables with impregnated paper insulation. This section
summarizes the main properties of this model.
Oil-paper insulation, in cables, is realized by wounding several paper foils one
around the other and by impregnating the foils with oil. The impregnation process cre-
ates a thin oil film between two consecutive paper foils, so that the whole structure can
be thought as an alternated sequence of paper and oil circular layers. As a consequence
of several chemical and physical processes, that can be ascribed to the presence of im-
purities in the insulation, [24], a non-negligible number of charged particles is found to
be dissolved in the oil. Under the application of a DC electric field the charged particles
move through the various layers of oil and (porous) paper by means of a hopping mech-
anisms, as suggested by Jeroense and Kreuger in [24]. This hopping mechanism may
be the reason of the observed conduction current in oil-paper insulation, that is typically
modelled with a temperature- and field-dependent electric conductivity [14, 23–25]:
 
α
σ(T, E) = σ0 exp − + β|E| , (2.24)
T
where σ0 is the reference conductivity at infinite temperature and no field, E is the elec-
tric field magnitude, α and β are, respectively, the temperature and field coefficients.
The temperature-dependence in (2.24) is the same as the one in (2.10), with α = W/kB .
The conductivity model (2.24) is empirical and the values of α and β are fitted from
measurements of the DC conduction current.
Occhini and Maschio [23] also suggested that not all the charged particles, that can
be of different types and sizes, travel across the entire insulation. Some of them, in fact,
may gather at the interfaces between the oil and the paper foils, giving rise to interfacial
polarization phenomena. Moreover, the oil is populated by polar molecules that, under

12
 2.2. The Occhini Model

the application of a DC field, rotate in such a way to align their electric dipole to
the direction of the field, which is the way dipolar polarization works. Atomic and
electronic polarization can be thought to take place both in oil and paper. In conclusion,
different types of polarization processes occur in impregnated paper, and the Debye
extended model can be adopted to describe the observed polarization current [23].
The Occhini model is an electric circuit that describes the transient current observed
in impregnated paper insulation under DC voltage, as shown in Figure 2.2. The con-
ductance G is related to the conductivity (2.24), the capacitance C∞ is related to the
fast polarization processes and the three R-C branches are associated to the slow polar-
ization processes. In general, more than three R-C branches can be put into the model,
but three of them are usually sufficient to describe most of the observed transients.
In order to work with the experiments, it is necessary to find the mathematical rela-
tion between the circuital parameters of the circuit and the microscopic parameters of
the classical theory of dielectrics, namely conductivities, susceptibilities and permittiv-
ities. First of all, recall the expression of the measured current, (2.9), that was written
as the sum between conduction and polarization currents. Using (2.1), (2.6) and (2.8)
yields
dD(t)
im (t) = ic (t) + ip (t) = AσE(t) + A . (2.25)
dt
It should be noted at this point that the model in Figure 2.2 is suitable to describe
the transient current in a flat specimen of oil-paper insulation as the one described in
section 2.1 [23]. To describe the actual cable insulation it is necessary to stack in series
a number of circuits like the one in Figure 2.2, but this matter will be discussed with
more detail in section 2.4. For the moment, the discussion will focus on a flat specimen
with area A and thickness d, placed between a high voltage electrode (represented by
node a in Figure 2.2) and a ground electrode (node b in Figure 2.2). In dielectric testing
the samples are always prepared in such a way that d2 ≪ A, which is an empirical
condition that guarantees, to a certain extent, a uniform electric field in the bulk of the
sample. Assuming that this condition holds, the electric field can be written as
V H(t)
E(t) = , (2.26)
d
being V H(t) the step voltage applied between nodes a and b. The displacement field
then is
V
D(t) = ε(t) H(t), (2.27)
d
which implies, using (2.23), that
M
dD(t) V X ε0 χk −t/τk V
= ε∞ δ(t) + e H(t), (2.28)
dt d k=1
τ k d

being δ(t) the Dirac delta function. Equation (2.28) has been obtained considering that
M
X  V
ε0 χk 1 − e−t/τk δ(t) = 0 ∀t ≥ 0, (2.29)
k=1
d

which is a term that appears in the computation of the time derivative of (2.27) . Finally,

13
Chapter 2. Polarization Processes and Electric Field Transient

R1 R2 R3
G C∞
C1 C2 C3

Figure 2.2: The Occhini model for a flat specimen of dielectric sample with M = 3 polarization pro-
cesses.

substituting (2.26) and (2.22) into (2.25), and defining

A
G=σ , (2.30)
d
A
C ∞ = ε∞ , (2.31)
d
A
C k = ε0 χ k , (2.32)
d
Ck
Gk = , (2.33)
τk
the measured current reads
M
X Ck V
im (t) = GV H(t) + C∞ V δ(t) + e−t/τk V H(t), (2.34)
k=1
τk

that corresponds to the current flowing between nodes a and b in the circuit of Figure 2.2
upon the application of the step voltage V H(t). Equation (2.34) highlights that the fast
polarization processes give rise to a surge current, C∞ V δ(t), that vanishes immediately
after the application of the voltage. Indeed, for t > 0, the measured current is
M
X Ck V
im (t) = GV + e−t/τk , (2.35)
k=1
τk

which means that during the transient only the current associated to the conduction
mechanisms and to the slow polarization processes is observed.

2.3 Fitting the Occhini Model Parameters from PDC Measurements

The parameters of the Occhini model are fitted from measurement of the so-called
polarization and depolarization currents (PDC). Using the experimental setup of Fig-
ure 2.1, a DC voltage is applied to the sample for a certain amount of time, tpol , until the
observed transient current reaches the DC steady state. Then the voltage is removed, the
electrodes are short-circuited and the resulting discharge current is measured. Under
applied voltage the measured current is the sum of the conduction, ic (t), and the po-
larization, ip (t), currents. After the voltage is removed the measured current coincides
with the so-called depolarization current, id (t). The situation is qualitatively described
by the diagram of Figure 2.3. The interpretation of the current curves can be carried on

14
 2.3. Fitting the Occhini Model Parameters from PDC Measurements

Polarization Depolarization
transient transient
Current
ic(t) + ip(t)

0 tpol Time
id(t)

Figure 2.3: Qualitative representation of the current measured from PDC experiments performed on flat
dielectric specimens.

the basis of the Occhini model in Figure 2.2. The initial current spike, observed both
at t = 0 and t = tpol , is due to the capacitance C∞ . It should be noted that in real
experiments the voltage source is never an ideal step, but the DC voltage is usually ob-
tained via a ramp or an exponential transient. The implication is that it is not possible to
obtain the value of C∞ from the PDC measurements. The value of C∞ , or equivalently
of ε∞ , is usually retrieved from the frequency response of the material, by assigning
to ε∞ the value of the real part of the complex permittivity at 1 Hz or 50 Hz. During
the polarization transient the measured current is given by (2.35). At DC steady state
the current in the R-C branches is negligible, therefore the value of the conductance is
calculated as
Is
G= , (2.36)
V
being Is the steady-state current. During the depolarization transient there is no applied
voltage, and the depolarization current is given by
M
X
id (t) = − Ak e−(t−tpol )/τk , (2.37)
k=1

with
Ck V  
Ak = 1 − e−tpol /τk . (2.38)
τk
Current id (t) is the discharge current of the capacitors in the R-C branches of the
Occhini model, that have been charging during the polarization transient. The term
1 − exp(−tpol /τk ) in (2.38) takes into account the fact that the capacitors may not be
completely charged when the voltage is removed at t = tpol . The depolarization current
is important because it does not include the contribution of the conduction processes. In
other words, id (t) represents the pure manifestation of the polarization processes taking
place in the dielectric, therefore it retains the relevant information to fit the parameters
of the R-C branches.
The problem of fitting the R-C parameters from the depolarization current can be
tackled in several ways. In [27–29] der Houhanessian and Zaengl proposed an ap-
proach to interpolate the measured depolarization current with a large number of R-C
branches. A set of regularly spaced τ -s was initially selected, spanning over the time
range of the measured current (usually from 1 s up to 104 s). The capacitances cor-
responding to the selected τ -s were calculated so to interpolate the measured current.

15
Chapter 2. Polarization Processes and Electric Field Transient

Measured
i1 (t)
10−4 t = t̄1 i2 (t)

Current [A]
i1 (t) + i2 (t)

t = t̄2
10−5

10−6
101 102 103 104
Time [s]

Figure 2.4: Qualitative representation of the way two exponential components of the depolarization
current (i1 (t) and i2 (t)) are combined together to represent the measured current.

After the first interpolation some of the branches had negative capacitance, hence they
were removed, and the interpolation was repeated until all the branches had positive ca-
pacitance. The problems with this approach are that the resulting number of branches is
high (∼ 10−100) and it depends on the initial choice of the τ -s. This is in contrast with
the observation of Occhini and Maschio [23] that 3 branches should, usually, be suffi-
cient to fit the depolarization current. Other authors proposed other techniques, relying
on vector fitting [30] or genetic algorithms [31], which seem unnecessarily complicated
and still retain the problem of yielding many R-C branches. The method adopted in this
work, that will be described now, was developed starting from the approach proposed
by der Houhanessian and Zaengl, with the aim of simplifying the procedure and reduc-
ing the number of R-C branches. The methodology has been published by the author of
this work in [15]. The technique is illustrated with the help of Figure 2.4, that shows a
qualitative idea of how two currents of the R-C branches, i1 (t) and i2 (t), are combined
to approximate the behaviour of the measured depolarization current (the black line).
The plot is in log-log scale, which is the typical one adopted for PDC measurements.
The idea behind the proposed fitting method is that the depolarization current ob-
served from the PDC measurements is often described by a power law like t−m , with
m > 0 [14]. Power laws of this kind are straight lines with slope −m in log-log plots of
current versus time. Consider the depolarization current component associated to the
k-th R-C branch, obtained from (2.37) by shifting, for the sake of simplicity, the origin
of the depolarization transient to t = 0:
ik (t) = Ak e−t/τk . (2.39)
Its natural logarithm, Ik , can be written as
t eξ
Ik = ln ik (t) = ln Ak − = ln Ak − , (2.40)
τk τk
with ξ = ln t. This implies that the derivative of ln ik (t) with respect to ln t, which is
the slope of the curve in log-log scale, is
d(ln ik (t)) dIk eξ t
= =− =− . (2.41)
d(ln t) dξ τk τk
16
 2.3. Fitting the Occhini Model Parameters from PDC Measurements

If (2.41) is evaluated in a certain time instant t̄k and the measured depolarization current
has slope −m in log-log scale, it is possible to estimate the time constant τk as
d(ln ik (t)) t̄k t̄k
=− = −m ⇔ τk = . (2.42)
d(ln t) t̄k τk m

The time instants t̄k are arbitrary. After M time instants t̄k have been manually selected
and the corresponding time constants have been estimated with (2.42), the condition
M
X
|idep (t̄k )| = ij (t̄k ) = idep,m (t̄k ) k = 1, . . . , M, (2.43)
j=1

being idep,m (t̄k ) the measured depolarization current at t = t̄k , can be exploited to
calculate the values of R-C branches capacitances. Indeed, using (2.39) and (2.38), it
follows that
M
X M
X M
X
−
ij (t̄k ) = Aj e t̄k /τk = akj Cj k = 1, . . . , M, (2.44)
j=1 j=1 j=1

with
V  −tpol /τj

akj = 1−e e−t̄k /τj . (2.45)
τj
Combining (2.43) and (2.44) yields the M × M linear system of equations
    
a11 . . . a1M C1 idep,m (t̄1 )
 . .. 
...   ...  =  ..
   
 .. . . . (2.46)
    
aM 1 . . . aM M CM idep,m (t̄M )

The method has been applied in [16] to fit a set of PDC measurements in impreg-
nated paper specimens. Three time instants t̄k were manually selected to have a first
interpolation, then a non-linear least squares optimization was run to obtain the best
fit. Table 2.1 summarizes the testing conditions and the properties of the samples in
four different experiments, that have been taken from a series of experiments described
in [32–34]. The measurements were performed on flat specimens with different areas
and thicknesses, and at different temperatures and applied electric fields. Samples S1
and S2 are taken from [32]. They were realized with a single layer of impregnated
paper, and a small electric field was applied (0.33 kV/mm). Sample S1 was tested

Table 2.1: Measurements conditions and properties of the samples. The electric field in the sample is
assumed to be uniform, thus the applied voltage, V , is related to the electric field as V = Ed. The
data has been taken from [32–34].

Sample Type T [◦ C] V [kV] E [kV/mm] A [cm2 ] d [mm]

S1 Single layer 30 0.5 0.33 141 1.5
S2 Single layer 60 0.5 0.33 141 1.5
S3 Multilayer 20 24 30 72 0.8
S4 Multilayer 50 30.5 12.2 78.5 2.5

17
Chapter 2. Polarization Processes and Electric Field Transient

at low temperature (30 [◦ C]), while S2 at high temperature (60 [◦ C]). Samples S3 and
S4 were made by several layers of impregnated paper stacked upon each other. S3 is
taken from [33] and it was tested at high electric field (30 kV/mm) and low tempera-
ture (20 [◦ C]). S4 comes from [34] and it was tested at high electric field (12.2 [◦ C]) and
high temperature (50 [◦ C]). The peculiarity of this set of experiments is that it covers
all the possible working conditions that are encountered in practice for impregnated
paper, namely high and low temperatures, high and low electric fields. If the Occhini
circuit is robust, then it must be able to model all these experiments while satisfying
the following constraints:
1. it should use a minimum number of R-C branches to fit the experiments (ideally
3), that consider polarization mechanisms involving the orientation of moieties
(dipoles) of different nature, and interface phenomena,
2. if the R-C branches are fitted from the depolarization current, the same parameters
must provide a polarization curve that agrees with the experiments.
The first constraint implies that three branches allow to identify the most relevant po-
larization processes and to divide them into faster, intermediate, and longer processes,
according to their time constants τk = Rk Ck . This is supported by Table 2.2, which
shows the time constants fitted from the depolarization currents shown in Figure 2.5.
Three branches are sufficient to reproduce the curves in the range 10–104 s, which is
that relevant to these measurements. Figure 2.5 compares the measured depolarization
curves (“msr”) of the samples, with the fitted ones (“fit”). Tables 2.3 and 2.4 report the
fitted values for Ck and Rk .
Table 2.2: Time constants fitted from the measurements in [32–34]. The results are from [16].

τk [s] S1 S2 S3 S4
τ1 2.18 · 101 2.70 · 101 3.62 · 101 4.01 · 101
τ2 2.19 · 102 2.87 · 102 3.00 · 102 4.19 · 102
τ3 2.57 · 103 1.24 · 103 3.10 · 103 2.84 · 103

Table 2.3: Capacitances fitted from the measurements in [32–34]. The results are from [16].

Ck [F] S1 S2 S3 S4
C1 2.56 · 10−11 1.95 · 10−10 1.70 · 10−11 1.41 · 10−11
C2 6.41 · 10−11 4.94 · 10−10 3.10 · 10−11 4.83 · 10−11
C3 1.84 · 10−10 3.90 · 10−11 1.28 · 10−10 6.76 · 10−11

Table 2.4: Resistances fitted from the measurements in [32–34]. The results are from [16].

Rk [Ω] S1 S2 S3 S4
R1 8.47 · 1011 1.38 · 1011 2.12 · 1012 2.841012
R2 3.42 · 1012 5.84 · 1011 9.67 · 1012 8.681012
R3 1.40 · 1013 3.20 · 1013 2.43 · 1013 4.191013

18
 2.3. Fitting the Occhini Model Parameters from PDC Measurements

S1 (msr) S3 (msr) S1 (fit) S3 (fit)

S2 (msr) S4 (msr) S2 (fit) S4 (fit)

10-08

10-09
Current [A]

10-10

10-11

10-12

102 103 104

Time [s]

Figure 2.5: Depolarization current measured (“msr”) in samples S1, S2, S3 and S4, and their corre-
sponding fittings (“fit”). The measurements are from [32–34], the results are from [16].

S1 (msr) S3 (msr) S1 (fit) S3 (fit)

S2 (msr) S4 (msr) S2 (fit) S4 (fit)

10-08
Current [A]

10-09

10-10

102 103 104

Time [s]

Figure 2.6: Polarization current measured (“msr”) in samples S1, S2, S3 and S4, and their correspond-
ing fittings (“fit”). The measurements are from [32–34], the results are from [16].

19
Chapter 2. Polarization Processes and Electric Field Transient

The remaining parameters, G and C∞ , necessary to complete the model, are listed
in Table 2.5: G was obtained from (2.36), while only the capacitance of sample S3 was
available from [32] (this is, however, not an issue, because as already observed from
(2.35) only the currents of the R-C branches and the conduction current are contribute
to the transient).
Table 2.5: DC conductances and high frequency capacitances fitted from the measurements in [32–34].
The results are from [16].

S1 S2 S3 S4
G [Ω] 6.33 · 10−14 1.22 · 10−12 9.09 · 10−14 1.03 · 10−13
C∞ [F] – – 2.79 · 10−10 –

Figure 2.6 shows the comparison between the experimental and fitted currents dur-
ing the polarization transient. It should be noted that the currents in Figure 2.6 are
called polarization currents, even though they include both the contribution of the po-
larization and conduction processes. This is a common nomenclature used in literature
and it will be used in this thesis as well, besides that fact that is not completely precise.
In conclusion, the agreement between experiments and fittings shown by Figure 2.5
and Figure 2.6 confirms the robustness of the Occhini model in reconstructing the po-
larization and depolarization currents in impregnated paper samples.

2.4 Equivalency between the Occhini Model and the Finite Difference
Discretization of the Cable Geometry
The previous section showed the effectiveness of the Occhini model to describe the
transient currents in thin flat samples of impregnated paper insulation. The next step
is to apply the model to the cable geometry and simulate the electric field transient in
the insulation layer. The simplest idea to perform this geometric transition is to divide,
conceptually, the cable insulation in a number of thin concentric circular layers, and
to associate to each of them an equivalent circuit like the one in Figure 2.2, with suit-
ably scaled circuital parameters. The resulting model is a series of Occhini equivalent
circuits stacked in series. This is the idea followed by Occhini and Maschio in [23].
Moreover, in [23] the approach was also validated experimentally. A number of alu-
minum foils were wound at regular intervals in the insulation of a suitably–prepared
cable insulated with impregnated paper. The aluminum was inserted between two pa-
per foils and served for the measurement of the electric potential at a specific radial
distance. The average electric field between two consecutive aluminum foils was es-
timated as the ratio between the potential drop and the distance between them. The
results reported in [23] show a good agreement between the measured transient electric
field and the theoretical calculations.
The approach followed by Occhini and Maschio works for transient field simula-
tions under the hypothesis of radial symmetry. The electric field can be assumed to
be radial in most situations, but if more specific simulations are to be performed, for
example by considering the presence of cavities or imperfections in the geometry that
lead to non-radial field distributions, the approach followed in [23] is no longer viable.
To this aim, numerical schemes like the finite difference method or the finite element

20
2.4. Equivalency between the Occhini Model and the Finite Difference Discretization of
the Cable Geometry

method are much more powerful, as they are not limited to geometries with radial sym-
metry. The purpose of this section is to show that the approach presented in [23] is
equivalent to the application of a finite difference scheme for the discretization of the
cable geometry under the assumption of radial symmetry. Since the finite difference
scheme can be easily generalized to non-radial 2D geometries, or even to 3D geome-
tries, the equivalency provides the logical bridge to generalize the Occhini model in
Figure 2.2 to any geometric configuration. The discussion is presented in the following
two subsections, where the finite difference discretization is developed first, and the
equivalency is proved next. This work has been published in [16].

2.4.1 Finite Difference Scheme for Semi-Discretization in Space

An insulation system under DC excitation can be described using the following quasi-
electrostatic model, which stems from the theory presented in section 2.1:


 ρ=∇·D


 ∂ρ

+∇·J=0 . (2.47)
 ∂t


 ∂Pk
 τk
 + Pk = ε0 χk E k = 1, . . . , M
∂t
The first two equations in (2.47) are the Gauss’ law for the electric displacement field
D and the continuity equation, being ρ the electric charge density and J the conduction
current density. The last equation is the Debye model for the polarization processes,
being Pk the polarization vector associated to the k-th process and E the electric field.
Recalling that
X M
D = ε∞ E + Pk , (2.48)
k=1
that the electric field can be written in terms of an electric scalar potential φ, E = −∇φ,
and that the conduction current density is related to the electric field via J = σ(T, E)E,
where the conductivity is given by (2.24), (2.47) can be written as:
  
PM
ρ = ∇ · −ε∞ ∇φ + k=1 Pk







 ∂ρ

− ∇ · σ(T, E)∇φ = 0 . (2.49)

 ∂t


 ∂Pk
 τk + Pk = −ε0 χk ∇φ k = 1, . . . , M


∂t
Under the hypothesis of radial symmetry ∇φ = ∂r φur and Pk = Pk ur , thus (2.49)
takes the form
  
1 ∂ ∂φ PM
−ε∞ r


 ρ= + r k=1 Pk



 r ∂r ∂r
 
∂ρ 1 ∂ ∂φ

− σr =0 . (2.50)


 ∂t r ∂r ∂r
∂Pk ∂φ


+ Pk = −ε0 χk

 τk
 k = 1, . . . , M
∂t ∂r
21
Chapter 2. Polarization Processes and Electric Field Transient

Before proceeding with the discretization of (2.50) it is convenient to transform each

equation by means of a suitable scaling and change of variables. Substituting the first
equation of (2.50) into the second, the system of equations is reduced to
 "  #  
 ∂ 1 ∂ ∂φ P M 1 ∂ ∂φ
 ∂t r ∂r −ε∞ r ∂r + r k=1 Pk − σr =0



r ∂r ∂r
(2.51)
∂Pk ∂φ



 τk
 = −Pk − ε0 χk k = 1, . . . , M
∂t ∂r
Swapping the time and space differentiation operators and regrouping yields:
 "   #
 1 ∂ ∂ ∂φ P M ∂φ
 r ∂r ∂t −ε∞ r ∂r + r k=1 Pk − σr ∂r = 0



(2.52)
∂Pk ∂φ



 τk
 = −Pk − ε0 χk k = 1, . . . , M
∂t ∂r
Introducing the following auxiliary variables
∂φ
Q∞ = −2πε∞ Lr , (2.53)
∂r
Qk = 2πLrPk , (2.54)
where L is length of the cable, and the following coefficients,
α = 2πε∞ Lr, (2.55)
βk = 2πε0 χk Lr, (2.56)
γ = 2πLrσ, (2.57)
the following system of equations is obtained,

∂φ

 Q∞ = −α
∂r





 ∂Qk ∂φ
τk = −Qk − βk k = 1, . . . , M , (2.58)

 ∂t ∂r
  

 ∂ ∂  PM  ∂φ
Q∞ + k=1 Qk − γ =0


∂r ∂t ∂r


which needs to be solved in the interval I = {rmin ≤ r ≤ rmax } for time t < T .
It is possible now to proceed with the semi-discretization of (2.51) in space, using
suitable centred finite difference formulas as follows. Interval I is partitioned into N
subintervals by means of N + 1 grid nodes, r(0) = rmin < r(1) < · · · < r(i) < · · · <
r(N ) = rmax , and the subintervals lengths and midpoints are defined as (see Figure 2.7):
∆r(i) = r(i+1) − r(i) , (2.59)
r(i) + r(i+1)
r̂(i) = , (2.60)
2
∆r̂(i) = r̂(i) − r̂(i−1) . (2.61)

22
2.4. Equivalency between the Occhini Model and the Finite Difference Discretization of
the Cable Geometry

L
r(i)
Δr(i)

r(i-1)
. .
r(i-1)

r(i)
r(i)

r(i+1)
r

Δr(i)
Figure 2.7: Discretization of the HVDC cable insulation.

The coefficients α, βk and γ are collocated at the midpoints by letting

α(i) = 2πε∞ Lr̂(i) , (2.62)
(i) (i)
βk = 2πε0 χk Lr̂ , (2.63)
 
γ (i) = 2πLr̂(i) σ r̂(i) . (2.64)

Assuming f (r) to be a smooth function

  over the entire
 interval
 I and denoting its nodal
(i)
values over the grid as f = f r (i) ˆ(i) (i)
and f = f r̂ , let

∂f  f (i+1) − f (i)

r(i) = , (2.65)
∂r ∆r(i)
fˆ(i+1) − fˆ(i)
  
∂f
r̂(i) = , (2.66)
∂r ∆r̂(i)
and the application of such formulas to (2.58) yields

(i) α  (i+1) (i)

Q∞ = − (i) φ −φ


∆r





 (i)

 (i) (i) βk  (i+1) (i)

 τk Q̇k = −Qk − ∆r(i) φ −φ k = 1, . . . , M



    , (2.67)
1 (i) PM (i) (i−1) PM (i−1)
Q̇∞ + k=1 Q̇k − Q̇∞ + k=1 Q̇k −


∆r̂(i)



 #

     
 γ (i) γ (i−1)
φ(i+1) − φ(i) − ∆r φ(i) − φ(i−1) =0


∆r(i) (i−1)




Introducing the following capacitances and conductances,

(i) α(i)
C∞ = , (2.68)
∆r(i)
(i)
(i) βk
Ck = , (2.69)
∆r(i)
γ (i)
G(i) = , (2.70)
∆r(i)

23
Chapter 2. Polarization Processes and Electric Field Transient

the system of equations (2.67) becomes

  
(i) (i) (i+1) (i)



 Q ∞ = −C ∞ φ − φ


 (i)
 
(i) (i)
 (i+1) (i)
τ Q̇ = −Q − C φ − φ k = 1, . . . , M

 k k

k k
    . (2.71)
(i) PM (i) (i−1) PM (i−1)
Q̇∞ + k=1 Q̇k − Q̇∞ + k=1 Q̇k −





     

(i) (i+1) (i) (i−1) (i) (i−1)
G φ −φ −G φ −φ =0





Equation (2.71) constitutes a system of differential-algebraic equations (DAE) that can

be solved via suitable time-stepping methods, the most commonly used being the back-
ward difference formulas.

2.4.2 Equivalence between the Semi-Discretized Model and a Sequence of Oc-

chini Circuits
This section shows the equivalence of the semi-discretized system of equations (2.71)
and the Occhini circuit, with suitable identification of the model parameters. The im-
portance of such an identification is twofold: on one hand it allows to determine the
physical parameters of the insulation model (2.49) from charging and discharging cur-
rent measurements of thin samples as in [15, 23], while, on the other hand, it allows
to perform simulations of non-flat and non-uniform geometries, such as the insulation
layer of a DC cable with non-uniform temperature distribution.
Consider the circuit of Figure 2.8. The Kirchhoff current law at node i yields
   
XM XM
Q̇(i) (i) (i−1)
∞ + Q̇k  − Q̇(i−1)
∞ + Q̇k  −
k=1 k=1

    
(i) (i+1) (i) (i−1) (i) (i−1)
G φ −φ −G φ −φ = 0. (2.72)

Defining the capacitance charges as

 
Q(i)
∞ = −C (i)
∞ φ(i+1)
− φ(i)
, (2.73)
 
(i) (i) (i)
Qk = −Ck φ(i+1) − φk , (2.74)
(i) (i)
and solving for the nodal voltages φk in terms of the branch currents Q̇k yields
 
(i) (i) (i) i (i) (i+1) (i)
Rk Ck Q̇k = −Qk − Ck φ −φ . (2.75)

Equations (2.73) and (2.72) are equivalent to the first and the third equations in (2.71),
respectively. Equation (2.75) is equivalent to the second equation in (2.71) if the resis-
(i)
tance Gk is defined as
(i) 1 τk
Rk = (i) = (i) . (2.76)
Gk Ck
24
2.5. Impact of Polarization Processes on the Electric Field Transient in Paper-Insulated
DC Cables

φ(i-1)

(i-1) (i-1) (i-1)

R1 R2 R3
(i-1) (i-1) .
φ1(i-1) Q (i-1) φ2(i-1) Q. (i-1) φ3(i-1) Q. (i-1)
G C∞ 1 2 3
. (i-1) C1
(i-1)
C2
(i-1) (i-1)
C3
Q∞

φ(i)

(i) (i) (i)

R1 R2 R3
(i) (i) φ1 (i) . (i) .
φ2(i) Q (i)
.
φ3(i) Q (i)
G C∞ Q1 2 3
. (i) (i) (i) (i)
Q∞ C1 C2 C3

φ(i+1)
Figure 2.8: Series of Occhini models stacked in series with M = 3 polarization processes.

In conclusion, Table 2.6 compares the expressions for the conductivities and the
capacitances derived in this section, which refer to the axisymmetric geometry of the
cable, and the ones of the flat sample provided in section 2.2, which refer to a planar
geometry. One can note that the physical parameters ε∞ , σ, χk and τk are the same,
while the geometric factors change.
Table 2.6: Relation of circuit elements to material physical parameters of a flat sample with area A and
thickness d, and a thin layer of axisymmetric cable insulation of thickness ∆r at distance r from the
cable axis.

Parameter Cable Flat sample

Lr
G 2π ∆r σ(r) σ Ar
Lr
C∞ 2π ∆r ε∞ ε∞ Ar
Lr
Ck 2π ∆r ε0 χk ε0 χk Ar
τk τk
Rk Ck Ck

2.5 Impact of Polarization Processes on the Electric Field Transient in

Paper-Insulated DC Cables
The objective of this section is to show how polarization processes affect the electric
field transient in HVDC cables. The most appropriate context is the one of paper-
insulated cables: indeed, it has been shown in section 2.3 that impregnated paper is well
described by the Occhini model, meaning that polarization processes may play a central
role in behaviour of the material. The results presented in this section summarize the

25
Chapter 2. Polarization Processes and Electric Field Transient

work published in [15] and [16].

The reference object is a HVDC cable with nominal voltage Vn = 300 kV, conductor
radius of 23 mm and insulation thickness d = 20 mm. Two different cable insulations
are considered: one is constituted by the material of sample S3, the other by the material
of S4, discussed in section 2.3. The electrical conductivity for both materials is assumed
to be only temperature-dependent:
 !
1 1 
σ(T ) = σ0 exp −α − . (2.77)
T Tref

The same α is adopted for both materials, using a typical value for impregnated paper
[23], while σ0 = σ(Tref ) is computed from (2.30), using the conductances of Table 2.5.
The parameters to model the conductivity of S3 and S4 are summarized in Table 2.7.
The value of Tref is taken from Table 2.1. Note that both T and Tref in (34) must be in
kelvin.
Table 2.7: Conductivity model parameters for materials S3 and S4.

Material α [K] σ0 [S/m] Tref [◦ C]

S3 1.01 · 104 1.01 · 10−14 20
S4 1.01 · 104 3.28 · 10−14 50

As for the dielectric properties of the two materials some assumptions were required
since not all the necessary data was available. First, the permittivity ε∞ associated to the
fast polarization processes is assumed to be the same for both S3 and S4. Capacitance
C∞ = 2.79 · 10−10 F of S3 is given in Table 2.5, and, through (2.31), it corresponds to
ε∞ = 3.5ε0 . Such a value is typical of impregnated paper, and it is used also for S4.
Susceptibilities χk were retrieved from the Ck via (2.32). Table 8 lists the χk obtained
for the two materials.
Table 2.8: Susceptibilites of materials S3 and S4.

Material χ1 χ2 χ3
S3 0.21 0.39 1.6
S4 0.51 1.7 2.4

Lacking measurements of polarization and depolarization currents at different tem-

peratures, it was assumed that, except for the conductivity, all the other properties of
S3 and S4 were temperature independent. In the following discussion, C1 is a cable
that uses S3 as insulating material, and C2 is the same cable but with S4 as insulating
material. Both cables are simulated assuming the same temperature distribution across
the insulation, with Tint = 55 ◦ C and Text = 25 ◦ C. This also fixes the distribution of
the electric conductivity.
The simulated transient is made of three consecutive stages: energization (from
0 to Vn ), polarity inversion (from Vn to −Vn ) and discharge (from −Vn to 0). The
commutations occur, respectively, at 0 hours, 36 hours and 72 hours. The voltage step at
each commutation is modelled with a ramp, from 0 to Vn within 2 s and from Vn to −Vn

26
2.5. Impact of Polarization Processes on the Electric Field Transient in Paper-Insulated
DC Cables

within 2 s [35]. For each cable two simulations were performed. One simulation was
done using the quasi–electrostatic model (2.47), that includes the polarization processes
and that will be address as the polarization model. The other simulation was done using
what can be called the standard model, that does not include the polarization processes:

 ρ=∇·D

∂ρ . (2.78)

 +∇·J=0
∂t
Since the standard model is the polarization model without the Debye relaxation equa-
tions, the differences between the two models’ predictions have to be ascribed to the
dynamic of the polarization processes. In particular, to highlight the differences be-
tween such predictions during the transient, the field difference was defined as:
Epol (r, t) − Estd (r, t)
∆E(r, t) = , (2.79)
Eavg
where Epol (r, t) and Estd (r, t) are computed from (2.47) and (2.78), respectively, and
Eavg = Vn /d = 15 kV/mm is the average insulation electric field.
Figure 2.9 and Figure 2.10 report ∆E(r, t) for C1 and C2, respectively. In both cases
the largest differences occur immediately after each commutation, because Estd (r, t)
evolves faster than Epol (r, t), which is, in turn, slowed by the polarization dynamic. The
white regions in Figure 2.9 and Figure 2.10 correspond to the condition ∆E(r, t) = 0:
this means that the field distribution in each cable, computed by the two models, has
the same profile and the same time evolution. It is noticeable that ∆E(r, t) displays
considerably large values: in C1 the maximum value is 5 kV/mm, while in C2 it is 12
kV/mm. These values are reached, respectively, 0.7 hours and 4 hours after the polarity
inversion, which occurs 36 hours after the energization
Figure 2.11 shows the electric field profiles at the end of the energization transient,
before the polarity inversion. The fields were computed using the polarization model
(p), and the standard one (s), for both cables. As mentioned, such profiles correspond to
the steady-state resistive distribution, which is governed by the same equation in both
models:  
∇ · (σE) = 0 ⇔ ∇ · e−α/T E = 0. (2.80)
Since it was assumed the same α for both the insulation materials, and since both cables
have the same temperature distribution, the resistive profiles in Figure 2.11 are all the
same.
Figure 2.12 shows the field profiles of the two cables, computed again by the polar-
ization (p) and standard (s) models, 2 hours after the polarity inversion. The fact that,
at a given time instant, the field profiles differ from each other, is another indication
that the transient duration is different in each cable. Of course, the duration changes
also according to the model used for the simulation, as highlighted by Figure 2.9 and
Figure 2.10. Therefore, Figure 2.12 can be used to understand how fast the electric field
evolves for the insulating material and the model employed for the simulation. Ideally,
a precise comparison should be based on the calculation of a time constant τ that is rep-
resentative of the transient duration. Such a quantity can be, in fact, calculated for the
standard model, where τ is obtained as the ratio between permittivity and conductivity:

27
Chapter 2. Polarization Processes and Electric Field Transient

108 30 %

20 %

72 10 %
Time [hours]

0%

36 -10 %

-20 %

0 -30 %
5 10 15
Position in insulation [mm]
Figure 2.9: Field difference ∆E(r, t) for cable C1, from transient to steady state. The differences
between Epol and Estd persist for a few hours and the highest ones are reached after the polarity
inversion.

108 80 %

60 %

40 %
72
Time [hours]

20 %

0%

-20 %
36
-40 %

-60 %

0 -80 %
5 10 15
Position in insulation [mm]
Figure 2.10: Field difference ∆E(r, t) for cable C2, from transient to steady state. The differences
betweenEpol and Estd persist for much more time than in C1, and they are greater.

28
2.5. Impact of Polarization Processes on the Electric Field Transient in Paper-Insulated
DC Cables

C1 (p) C2 (p) C1 (s) C2 (s)

40
Electric field [kV/mm]

30

20

10

0
20 25 30 35 40 45
Position [mm]
Figure 2.11: Resistive field profiles at the end of the energization transient, and just before the polarity
inversion, computed using the polarization (p) and the standard model (s). All the curves are over-
lapped because, at steady state, all models satisfy equation (2.80).

C1 (p) C2 (p) C1 (s) C2 (s)

0
Electric field [kV/mm]

-10

-20

-30

-40
20 25 30 35 40 45
Position [mm]
Figure 2.12: Field profiles 2 hours after the polarity inversion, computed using the polarization (p) the
standard model (s).

29
Chapter 2. Polarization Processes and Electric Field Transient

40

Maximum electric field [kV/mm]

30

20 Polarity
Energization inversion Discharge

10
C1 (p) C1 (s)
C2 (p) C2 (s)

0
0 36 72 108
Time [hours]

Figure 2.13: Maximum electric fields in C1 and C2 during the simulated transient, computed using the
polarization model (p) and the standard one (s).

in this way it is possible to compare the time constants of C1 (s) and C2 (s). Consid-
ering, indeed, that both C1 (s) and C2 (s) are computed using the standard model, with
the same α and the same temperature profile, but with different Tref , it follows that:
τC2(s) ε∞ /σS4 σS3 e−α/Tref,S3
= = ≃ 7, (2.81)
τC1(s) ε∞ /σS3 σS4 e−α/Tref,S4
meaning that the electric field transient in C2 (s) is slower than in C1 (s). This is
coherent with Figure 2.12, where the profile of C1 (s) is already inverted with respect
to the starting configuration of Fig. 8, while the field in C2 (s) is still inverting. As for
the fields in C1 (p) and C2 (p), the analysis is more difficult because the polarization
model (3) involves more than one time constant: there is the τ calculated as ε/σ, but
also the time constants of the polarization processes. It is therefore impossible to derive
a single time constant that represents the field transient. On the other hand, it is possible
to make some qualitative considerations. Comparing C2 (s) and C2 (p) in Figure 2.12,
C2 (p) lags C1 (s). In the same way, C1 (p) lags C1 (s). This indicates that the field
evolution computed with the polarization model (p) is slower than the one computed
with the standard model (s). The polarization processes slow down the electric field
evolution in the cable.
Figure 2.13 shows the maximum electric field in the insulation of C1 and C2 during
the transient,
Em (t) = max |E(r, t)|, (2.82)
r

with E(r, t) being the field computed using either the polarization (Epol ) or the stan-
dard (Estd ) model. During the energization and the polarity inversion transients, before
reaching the steady state, Em (t) computed using the polarization model (p) is lower
than the one computed with the standard model (s), and this happens for both C1 and

30
2.5. Impact of Polarization Processes on the Electric Field Transient in Paper-Insulated
DC Cables

C2. During the discharge transient the opposite situation occurs. This finding high-
lights the very contribution of polarization processes to the field transient in HVDC
cables: under the application of a DC voltage, the maximum field intensity experienced
by the cable during the transient is lower in the presence of polarization processes. At
steady state the maximum field is the same for both models, as already derived from
equation (2.80). In conclusion, the cable works under reduced stress with respect to the
steady-state condition for longer time in the presence of polarization processes.

31
 CHAPTER 3
Modelling Issues in Polymers

This chapter discusses the inadequacy of Occhini’s model for the description
of the transient response of polymeric materials to step DC fields, that is
inevitably non-linear. Such an inadequacy warrants further studies of the
basic differences between materials and the development of more general
mathematical models, which will be carried out in the next chapter.

3.1 Non-Linear Response

In section 2.2 it was shown that the polarization and depolarization currents, (2.35) and
(2.37), can be written, according to the Occhini model, as
M
X Ck V
ip (t) = GV + e−t/τk , (3.1)
k=1
τk
and
M
X Ck V  
id (t) = 1 − e−tpol /τk e−t/τk , (3.2)
k=1
τk
where the expressions have been written with the same reference for time, t = 0, and
the depolarization current has been taken with positive sign for the sake of convenience.
If (3.2) is subtracted from (3.1), the result is a time-dependent quantity:
M
X Ck V
ip (t) − id (t) = GV + e−t/τk e−tpol /τk (3.3)
k=1
τk
However, assuming that the polarization time, tpol , is much larger than all the time
constants of the M polarization processes, tpol ≫ τk ∀k, the difference in the above

33
Chapter 3. Modelling Issues in Polymers

Figure 3.1: Polarization and depolarization currents measured in a specimen of XLPE at 60 ◦ C and
under an electric field of 20 kV/mm, and in sample S2 of impregnated paper, already introduced in
chapter 2.

expression can be considered to be by all means constant:

ip (t) − id (t) ≃ GV = const. (3.4)
It should be observed that G is related to the temperature- and field-dependent conduc-
tivity (2.24), but since the PDC experiments are performed under constant temperature
and field the conductance is constant too.
Consider now the polarization and depolarization currents measured in a specimen
of cross-linked polyethylene (XLPE) and in a sample of impregnated paper (S2), shown
in Figure 3.1. The difference between the two currents, |ip (t) − id (t)|, yields the result
of Figure 3.2: the difference is not constant in time for XLPE, contrarily to (3.4), while
it is constant for S2. This fact is an indication that the response of XLPE is not linear,
as suggested by Jonsher [24]. In particular, it implies that the Occhini model, which
is linear, is not suitable to describe the currents observed in the XLPE sample. On the
other hand, the response is linear for S2, but this fact was already confirmed in chapter
2. Following Jonsher, after the application of step DC field charges are injected form
the electrodes into the bulk of the polymeric material. Under the driving force of the
electric field the movement of the injected charges takes place on a significant scale.
The injection and the movement of charges are non-linear phenomena, and their effect
is enhanced at higher fields and temperatures [36].
A particularly clear manifestation of the non-linearity is the crossing over of the
polarization and depolarization currents in XLPE at short times, that is observed in
Figure 3.1. It was suggested by Jonsher and Careem, [36], that the injected charges
accumulate near one or both electrodes under the applied field. When the field is sud-
denly removed the injected charges are quickly depleted, giving rise to the large depo-
larization current current that is observed in the early instants of Figure 3.1 and that

34
 3.2. Inadequacy the Occhini Model for Polymers

Figure 3.2: Absolute value of the difference between the polarization and depolarization currents shown
in Figure 3.1, |ip (t) − id (t)|. The plot has been obtained by interpolating the curves of Figure 3.1 in
the range between 1 s and 4.5 · 103 s.

overcomes the polarization one.

As a last note, it should be observed that the presence of non-linear phenomena does
not mean that linear polarization processes no longer occur. They are still present, and
they evolve in parallel the other phenomena. Therefore, the Occhini model does not
have to be thrown away, but it is rather necessary to extend it in order to account for
the non-linearities: this matter will be treated in the next chapter. It should be also
noted, after this last consideration, that the currents in Figure 3.1 are not polarization
and depolarization ones in a strict sense, as they result from the combination of several
processes running in parallel. It would be better to refer to them as the charging current,
when the voltage is applied to the sample, and as the discharging current after the
voltage is removed. However, in the engineering jargon the terms polarization and
depolarization are used as synonyms for charging and discharging, so in this work they
will be used interchangeably, as long as the precise meaning will be clear from the
context.

3.2 Inadequacy the Occhini Model for Polymers

The previous section discussed the matter of non-linear phenomena and the inadequacy
of the Occhini model in describing them in a general sense. This section presents a
more specific case study, by attempting to fit the PDC measurements obtained from
a specimen of polypropylene (PP) with the Occhini model, using different strategies.
The failure of all the attempts will definitely prove the inadequacy of the Occhini model.
The content of this section has been published in [17].
The polypropylene sample was tested at 30 kV/mm and 60 ◦ C. Three different fit-
ting attempts were made:

35
Chapter 3. Modelling Issues in Polymers

1. the R-C branches were fitted from the depolarization current using the interpola-
tion technique described in section 2.3;
2. the R-C branches were fitted from the polarization current using the interpolation
technique described in section 2.3;
3. the R-C branches were fitted from the depolarization current using the interpola-
tion technique described by Houhanessian and Zaengl [27–29].
An attempt was considered valid if by fitting one of the measured currents it was pos-
sible to reconstruct the other one.

3.2.1 First Attempt: Fitting the Depolarization Current with 3 R-C Branches
The fitting of the depolarization current, depicted in Figure 3.3, has to consider two
separate time-ranges of the transient. The first range includes the early 20 s, in which
the trend of the polarization current resembles the half of a hump. This type of trend can
be very well represented by an exponential transient, that in the case of the measurement
of Figure 3.3 is due to the generator of the experimental setup. The voltage applied by
the DC generator can be represented by

vs (t) = V e−t/τg , (3.5)

with τg = 2.5 s. The voltage (3.5) causes a transient current in the capacitor C∞ of the
Occhini model, Figure 2.2, that is given by
dvs C∞ −t/τg
iC∞ (t) = C∞ =− e , (3.6)
dt τg

10 -06
10 -07
10 -08
10 -09
10 -10
Current [A]

10 -11
10 -12
10 -13
10 -14 Polarization current
10 -15 Depolarization current

10 -16 Polarization fi�ing

10 -17 Depolarization fi�ing

10 -18
10 -1 10 0 10 1 10 2 10 3 10 4 10 5 10 6
Time [s]

Figure 3.3: Fitting of PDC measurements in a polypropylene sample at 30 kV/mm and 60 ◦ C. The
fitting is based on the depolarization current and uses 3 R-C branches.

36
 3.2. Inadequacy the Occhini Model for Polymers

as well as a current in the conductance G,

iG (t) = Gvs (t) = GV e−t/τg . (3.7)
It was found that the current associated to the polarization processes is negligible in the
first 20 s, so that, in principle,
id (t) ≃ iG (t) + iC∞ (t) for t < 20 s. (3.8)
However, current iG (t) was found to be negligible during the depolarization, therefore
id (t) ≃ iC∞ (t) for t < 20 s, (3.9)
and the fitting of the first 20 s can be obtained by adjusting the value of C∞ . After the
first 20 s current iC∞ (t) drops rapidly to zero, the voltage source is practically at steady-
state, vs (t) ≃ v(t), thus the depolarization current is given by the usual expression
M
X Ck V  
id (t) ≃ − 1 − e−tpol /τk e−t/τk , (3.10)
k=1
τk
which legitimates the application of the fitting method described in section 2.3. This
separation of the transient in different ranges, each of which has a dominant current
component, is a typical approach followed in the fitting of experimental measurements,
especially when data is available form the very beginning of the transient. It must be
observed that the separation is guided by the model adopted to fit the measurements,
hence different models may lead to different ranges.
The polarization and depolarization currents measured in Figure 3.3 were fitted with
C∞ = 4 · 10−11 F and G = 1.7 · 10−14 S. The parameters of the R-C branches are listed
in Table 3.1.
Table 3.1: R-C branches fitted from Figure 3.3.

k Rk [Ω] Ck [F] τk [s]

1 1.56 · 1013 6.61 · 10−13 1.03 · 101
2 8.58 · 1014 1.40 · 10−13 1.20 · 102
3 9.96 · 1015 1.10 · 10−13 1.10 · 103

It is evident, from Figure 3.3, that while it is possible to obtain a very good fitting
of the depolarization current, the transient of the polarization current cannot be recon-
structed. Indeed, consider the ratio between the polarization current associated to the
k-th R-C branch, ik , and the steady-state conduction current:
Ck V −t/τk
ik τ
e 1 −t/τk
= k = e , (3.11)
GV GV Rk G
where 1/Rk = Ck /τk . The ratio 1/Rk G for each of the R − C branches is
1 1 1
≃ 4, ≃ 0.7, ≃ 0.006, (3.12)
R1 G R2 G R3 G
which means that the polarization currents associated to the longer time constants, τ2
and τ3 are too small compared to the steady-state current, hence they cannot provide
any significant contribution to the transient polarization current. In conclusion, fitting
the R-C branches from the depolarization current is not a viable solution.

37
Chapter 3. Modelling Issues in Polymers

3.2.2 Second Attempt: Fitting the Polarization Current with 3 R-C Branches
This attempt is based on the same considerations of the previous one, but this time the
R-C branches are fitted from the polarization current. The first 20 s of the transient are
still dominated by the capacitive current iC∞ (t), and the capacitance C∞ is the same.
The fitting of the current after 20 s is obtained by subtracting the steady-state current,
GV , from the measured one, so that the “true” polarization current should be observed:
ip (t) ≃ ip,m (t) − GV, (3.13)
where ip,m (t) is the measured polarization current. The value of G remains the same.
Table 3.2 summarizes the shows of the R-C branches fitted from (3.13).
Table 3.2: R-C branches fitted from Figure 3.4.

k Rk [Ω] Ck [F] τk [s]

1 1.62 · 1013 3.65 · 10−12 5.91 · 101
2 5.26 · 1014 2.61 · 10−11 1.37 · 102
3 6.18 · 1015 3.90 · 10−10 2.41 · 103

Figure 3.4 shows that the fitted parameters are not able to reproduce the measured
depolarization current. In particular, the depolarization current simulated using the
values in Table 3.2 is much larger than the measured one.

10 -06
10 -07
10 -08
10 -09
10 -10
Current [A]

10 -11
10 -12
10 -13 Polarization current
10 -14 Depolarization current
10 -15 Polarization fi�ing
10 -16 Depolarization fi�ing
-17
10
10 -1 10 0 10 1 10 2 10 3 10 4 10 5 10 6
Time [s]
Figure 3.4: Fitting of PDC measurements in a polypropylene sample at 30 kV/mm and 60 ◦ C. The
fitting is based on the polarization current and uses 3 R-C branches.

3.2.3 Third Attempt: Fitting the Depolarization Current with more than 3 R-C
Branches
One last attempt to fit the PDC measurements can be done by increasing the number
of R-C branches used to fit one of the currents. Starting from the depolarization cur-

38
 3.2. Inadequacy the Occhini Model for Polymers

rent, as done in the first attempt, the method by der Houhanessian and Zaengl [27–29]
suggests to select a large number of R-C branches whose time constants are uniformly
spanned in a range of values. To obtain the result of Figure 3.5, 500 time constants
were distributed between 3 s and 4 · 103 s. After an interpolation, the branches with
negative capacitances were eliminated, and the process was iterated until only branches
with positive capacitances were left. In the case of Figure 3.5 the processes ended up
with 7 branches, but, even if the fitting of the depolarization current is slightly better
than the case of Figure 3.3, the fitting of the polarization current is still not acceptable.

10 -06
10 -07
10 -08
10 -09
10 -10
Current [A]

10 -11
10 -12
10 -13 Polarization current
-14
10 Depolarization current
-15
10 Polarization fitting
-16
10 Depolarization fitting
-17
10
10 -1 10 0 10 1 10 2 10 3 10 4 10 5 10 6
Time [s]
Figure 3.5: Fitting of PDC measurements in a polypropylene sample at 30 kV/mm and 60 ◦ C. The
fitting is based on the polarization current and uses the der Houhanessian and Zaengl method [27–
29], that resulted in 7 R-C branches.

39
 CHAPTER 4
Trapping and Detrapping in Polymers: a Circuital
Model

This chapter presents a dynamical model that describes the charging and
discharging transients observed in flat samples of polymeric materials. After
a brief review of the band theory of solids, which constitutes the theoreti-
cal framework of this chapter, the model equations are derived under suit-
able simplifying assumptions. The theoretical presentation will result in an
equivalent electric circuit for the flat polymeric sample. The latest sections
present the experimental validation, obtained by fitting PDC measurements
performed in polypropylene and XLPE samples, and a discussion on the lim-
itations of the presented model.

4.1 Preliminary Considerations on the Model

Before diving into the details of the model, it is worth providing some context to the
problem of describing the physical behaviour of polymers and, more in general, dielec-
tric materials. The extensive study carried by Jonsher on a large variety of materials
led to the conclusion that the transient discharge current obtained after charging the
material with a steady DC field is described by the following empirical law [14]:
1
i(t) ∝ , (4.1)
(ωp t)n + (ωp t)1+m
where n, m ∈ [0, 1] and ωp > 0 are fitting parameters related to the specific material.
Expression (4.1) is reduced to the following power law for short times,
1
i(t) ∝ t−n for t ≪ , (4.2)
ωp
41
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

and to another power law for long times,

1
i(t) ∝ t−1−m for t ≫ . (4.3)
ωp

The dominance of two different power laws, t−n and t−1−m , suggests that two distinct
physical processes determine the material’s response at different time scales. A model
that aims at describing this behaviour must be able to reproduce (4.1), or at least of one
of the two power laws (4.2) or (4.3), depending on the time range that the model itself
and the available measurements allow to evaluate.
Once the “target output” has been established, the next step is to investigate the
physics that produces it. The previous chapter has shown that the Occhini model is able,
in fact, to approximate a power law, but it is not able, on the other hand, to describe
the full set of PDC measurements observed in XLPE. A different physical basis has to
be sought. The traditional view is that polymers may exhibit those physical processes
that are typically observed in semiconductors, such as hopping and diffusive transport,
trapping, detrapping, generation and recombination of charge carries, injection from
metallic contacts and modification of potential barriers. Under this perspective the band
theory of solids constitutes the theoretical framework to build models for polymeric
materials. It is interesting to note that some of these processes provide, under suitable
simplifying assumptions, an exponential response to a DC excitation, therefore they
may be represented by R-C branches like the Debye processes [14]. Of course the
exponential response is too restrictive to represent experiments such as those shown for
XLPE in Figure 3.1, hence looser hypotheses will be adopted.
On the use of band theory there is one crucial aspect that requires a brief mention.
Band theory is essentially a single-particle theory, in the sense that each charge carrier
(typically an electron or a hole) is a non-interacting particle that behaves independently
from the others. Therefore the total current arising from a certain material is simply the
sum of the contribution of each carrier. There is, however, reasonable suspicion that
charges in polymers may strongly interact with each other and that the power law (4.1)
emerges from all such interactions. This led to the development of theories to take
into account the mutual interactions, such as the work of Dissado and Hill [37] and
the many-body universal model by Jonscher [14]. With this concern in mind, even if
a model stemming from band theory provides an acceptable fitting to the experimental
results, one must be very careful in drawing definitive conclusions from it. For ex-
ample, PDC measurements are performed in specific ranges of temperature (e.g., from
20◦ C to 90◦ C) and average electric field (e.g, from 20 kV/mm to 100 kV/mm). Infer-
ence of the material behaviour outside of the observed temperature and field ranges is
dangerous, since the model may no longer be valid.

4.2 Band Theory of Solids: Key Concepts

This section recalls the key concepts of the band theory of solids that will be used to
derive the polymer model. The content is derived from Smith’s book [38].
Band theory describes the energy levels available for electrons in a solid, that deter-
mine the response of the material to different types of excitation. Some of the allowed
levels lie either below the upper limit of the valence band, Ev , or above the lower limit

42
 4.2. Band Theory of Solids: Key Concepts

E
Conduction band
Ec

Forbidden gap
EF ΔE

Ev
Valence band
Figure 4.1: The energy band diagram of a perfect non-metallic solid. Ec is the lower bound of the
conduction band, Ev is the upper bound of the valence band, EF is the Fermi level and ∆E =
Ec − Ev is the width of the forbidden gap.

of the conduction band, Ec , as illustrated by Figure 4.1. In a metallic solid the valence
and conduction bands are slightly overlapped, while in a non-metallic solid, a semicon-
ductor or an insulator, the valence band lies below the conduction band, Ev < Ec , and
the separation between them is called the forbidden gap, whose width is ∆E = Ec −Ev .
The probability of a certain energy level, E, to be occupied by an electron is given by
the Fermi-Dirac statistics, provided that the solid is in thermodynamic equilibrium,
1
f (E) =  , (4.4)
E−EF
1 + exp kB T

where EF is the Fermi energy, that corresponds to that energy in which the occupation
probability is 12 . The Fermi energy lies in the forbidden gap, not necessarily midway
between Ec and Ev . The probability of occupation by a hole, which is the probabil-
ity that an electron has been removed from a certain energy level, is 1 − f (E). The
qualitative behaviour of f (E) and 1 − f (E) is depicted in Figure 4.2. Assuming that
the material is subject only to thermal excitation, the total number of electrons in the
conduction band is given by
Z +∞
n= Nc (E)f (E) dE, (4.5)
−∞

where Nc (E) is the density of states (DOS) available in the conduction band for the
electrons at a certain energy E. For the present discussion the conduction band can
be assumed to be concentrated on a single energy level Ec (this, in particular, means

f(E) 1-f(E)
1

0.5

Ev E F Ec E
Figure 4.2: The probability density function of electrons, f (E), and holes, 1 − f (E).

43
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

neglecting the molecular disorder in a polymer), i.e.,

Nc (E) = Nc δ(E − Ec ). (4.6)

Moreover, the ratio (E
− EF )/kB T in (4.4) can be assumed to be ≫ 1, hence f (E) ≃

exp −(E − EF )/kB T . The density of electrons in the conduction band, then, is writ-
ten analytically as  
EF − Ec
n ≃ Nc exp (4.7)
kB T
Correspondingly, being Nv (E) be the DOS for holes in the valence band, the number
of holes is calculated as
Z +∞
 
p= Nv (E) 1 − f (E) dE, (4.8)
−∞

and it can be shown that, by measuring the energy in an opposite direction with respect
to the electrons and using the same arguments as before,
 
Ev − EF
p ≃ Nv exp . (4.9)
kB T
It is useful, for the following discussion, to introduce the concept of perfect non-
metallic solid, that is a solid with no allowed states in the forbidden gap. If g(E) is the
DOS in the forbidden gap, then g(E) = 0 for E ∈ (Ev , Ec ) in a perfect non-metallic
solid. This definition reflects the fact that, under the effect of thermal excitation only,
electrons and holes are always generated and destroyed in pairs, which implies that

ṅ = ṗ. (4.10)

Equation (4.10) determines the Fermi level in (4.4). The band structure of a perfect
non-metallic solid is, however, only an idealized model. In practice, there are allowed
energy levels in the forbidden gap that modify the band structure of Figure 4.1 and
hence the behaviour of the material. In the following the causes and consequences of
such modifications will be briefly and qualitatively described. For a more thorough
discussion on these topics, see [38].

4.2.1 Donors and Acceptors

A donor is an atom with an extra electron in its valence shell that does not take part
in the bonding with nearby atoms. The excess electron is thermally energized to the
conduction band with relative easiness with respect to the electrons in the valence band,
since the energy gap with the conduction band is much lower than the amplitude of
the forbidden gap, as shown in Figure 4.3. An acceptor has one electron less than is
required to form a normal bonding, hence it can easily acquire an electron from the
valence band, that leaves a hole behind (again, see Figure 4.3).
Donors and acceptors change the density of the charge carriers in the conduction
and valence bands with respect to the case of a perfect non-metallic solid. Under suit-
able simplifying assumptions [38], the effect is to either increase the electrons in the
conduction band and reduce the holes in the valence band, or reduce the electrons in the

44
 4.2. Band Theory of Solids: Key Concepts

E
_
Ec
+ + +
f '(E) Donors
EF'
EF
Acceptors
f(E) _ _ _
Ev +

0.5 1
Figure 4.3: The modified energy band diagram when donors and acceptors are introduced in a non-
metallic solid. f ′ (E) is the Fermi-Dirac distribution shifted from the one of the perfect non-metallic
solid, f (E). EF′ is the Fermi level of f ′ (E), while EF is the Fermi level of f (E).

conduction band and increase the holes in the valence band. The densities of electrons
and holes, n′ and p′ respectively, are related to the densities of a perfect non-metallic
solid, n and p, via
n′ p′ = np. (4.11)
The Fermi-Dirac statistics, f ′ (E), is shifted with respect to the statistics of the perfect
non-metallic solid, f (E), as shown by Figure 4.3. In particular this means that the
Fermi energy, EF , has moved to a different level, Ef′ .

4.2.2 Trapping Levels

Donors and acceptors introduce allowed energy levels in the forbidden gap, i.e. g(E) ̸=
0 for E ∈ (Ev , Ec ). Donors that have released an electron to the conduction band are
indeed able to capture another electron, either from the conduction band, the valence
band or an acceptor. In the same way, acceptors that have acquired an electron from
the valence band can always release it back, either to the valence band, the conduction
band or a donor. The release of an electron by an acceptor can be treated equivalently
as the capture of a hole. The capture of a carrier is called trapping, while the release
of a trapped carrier is called detrapping. The energy levels involved in these processes
are called trapping levels. The levels close to the conduction and valence bands are
indicated as shallow traps. This nomenclature is useful to distinguish them from deep
traps, that are energy levels originating from defects (in semiconductors) or micro-
scopic disorder (in polymers) that are far away from both the conduction and valence
bands. The energy levels in the forbidden gap can be in general single levels, like the
shallow traps on the left diagram of Figure 4.4, or a continuous distribution of levels,
as the deep traps depicted in the same diagram.
By providing levels that are midway between the conduction and valence bands,
deep traps enable the exchange of charge carriers between the two energy bands. The
right diagram of Figure 4.4 shows two simple ways in which this can happen. The gen-
eration of an electro-hole pair occurs when an electron is captured by a deep trap from
the valence band, and then it is excited to the conduction band. The recombination of
the electron-hole pair occurs when the electron is captured from the conduction band

45
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

E E _ _
Ec Ec
Shallow traps _ _

Recombination
Generation
Deep traps Et Deep trap

_ _
Shallow traps
Ev Ev
+

Figure 4.4: (Left) Shallow and deep traps in non-metallic solids. The traps can be either single energy
levels or a continuous distribution of them. In the picture, shallow traps are represented as single
energy levels, while deep traps are represented as a distribution of energy levels. (Right) Scheme
of the recombination and generation of an electron-hole pair taking place via a single deep trap at
energy Et . These processes occur also via more than one trap.

by the deep trap and then it falls back to the valence band. Generation and recombina-
tion then are a sequence of trapping and detrapping processes, that may also occur via
intermediate steps involving more traps.
Trapping levels have an asymmetric behaviour. They can capture a carrier with
relative easiness, but they are less prone to release it. The time required to free a
trapped carrier increases with the depth of the trap, so the release by deep traps takes
much more time than by shallow traps.

4.2.3 Surface Trapping Levels

The surface of a non-metallic solid can absorb impurities that act as traps for electrons
and holes. This results in the introduction of deep trap levels in the forbidden gap,
that are called surface trapping levels. Consider a non-metallic solid with n ≫ p,
i.e. in which most of the free carriers are electrons. Suppose that the surface trapping
levels are distributed both above and below the Fermi level, EF , as shown by Figure 4.5.
Electrons in the conduction band will tend to fill up the trap levels below EF , according
to the Fermi-Dirac statistics.
eϕs
Ec

Surface _
_
Electric field
+
EF
trapping levels _
_
+
+
+
_ +

Surface charge Depletion region

(negative) (positive charge)

Figure 4.5: Representation of the effect of surface trapping levels in a non-metallic solid. The depletion
region is drawn next to the surface charge to remark the fact that they are both close to the surface,
however the energy level of the depletion region is not necessarily the same of the surface charge.

As illustrated in Figure 4.5, the electrons captured by the surface trap levels are
located on the external side of the solid, forming a negative surface charge. This surface

46
 4.2. Band Theory of Solids: Key Concepts

charge will cause a positively charged depletion region inside the solid, from which all
the electrons are removed. The surface charge and the depletion region create a strong
electric field that raises the surface potential by an amount ϕs . The raised potential on
the surface causes a raise of the conduction band at the surface by an amount eϕs , being
e the absolute value of the electron’s charge.

4.2.4 Metallic Contacts

Contacts between a non-metallic solid and a metal give rise to phenomena that are de-
termined by the surface properties of the two objects. For this matter, a very important
property is the work function of a surface, which is defined as the minimum work re-
quired to remove an electron from a solid to a point in the vacuum that is still very close
to the surface. Mathematically, the work function is defined as

W = −eϕ − EF , (4.12)

where ϕ is the surface potential and EF is the Fermi energy. Typically, the work func-
tion of a metal is higher than the work function of a non-metallic solid [38]. The im-
plication is that, considering a non-metallic solid in which electrons are the major free
carrier, upon contact the electrons will flow into the metal, leaving a depletion region
in the non-metallic solid. The situation is qualitatively similar to Figure 4.5.

4.2.5 Application of an External Electric Field

The discussion carried so far presumed the absence of any external electric field. It is
now time to introduce the field and discuss its implications. For the sake of simplicity,
uniformity of the external electric field is assumed, as well as a uniform temperature
throughout the non-metallic solid. The external field is obtained by applying a potential
difference across a slab of the non-metallic solid. With the assumption of uniform
field the electrostatic potential, ϕ(x), decays linearly across the material’s volume. The
boundary Ec of the conduction band is related to the electrostatic potential by

Ec (x) = constant − eϕ(x), (4.13)

and a similar relationship, with a different constant, holds for Ev (x). The modified
structure of the energy diagram is depicted in Figure 4.6. Under the external electric
field electrons lying in the valence band or in the forbidden gap can get excited to the
conduction band, modifying the densities of carriers:

n′ = n + ∆n and p′ = p + ∆p. (4.14)

In this condition the non-metallic solid is no longer in thermodynamic equilibrium:

the Fermi-Dirac statistics is not valid. However, if the carriers’ concentration is not
far from the equilibrium (∆n ≪ n and ∆p ≪ p), electrons in the conduction band
can be considered in quasi equilibrium, and they can still be described by a modified
Fermi-Dirac statistics with the same functional form of (4.4), where two separate quasi
Fermi energies EF,n and EF,p are defined for electrons and holes, respectively, which
are different from the Fermi energy in thermal equilibrium, EF .

47
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

E Ec(x) _ _

ϕ>0
Electric
field

ϕ=0
Energy
level
Ev(x)

x=0 x=L
Figure 4.6: Modified energy band structure under an external electric field. The picture shows a slab of
non-metallic solid (the gray area) with thickness L.

4.2.6 Injection and Extraction

Carriers’ injection and extraction require both the contact with a metal and an exter-
nal electric field. The situation is depicted in Figure 4.7, where modifications of the
conduction band due to the scalar potential, depicted in Figure 4.6, has been omitted
for the sake of clarity. It was already shown in Figure 4.5 that surface states and the
contact with a metal create a depletion region at the interfaces, placed at x = 0 and
x = L in Figure 4.7. The electric field across the interfaces raises the conduction band
to a level that prevents any further electron to flow into the metal (Figure 4.7, above).
When a scalar potential, ϕ, is applied (Figure 4.7, below), the conduction band at the
interface at x = 0 is raised: electrons flow from the non-metallic solid to the metal
(extraction). At x = L the conduction band is lowered, allowing more electrons to flow
into the non-metallic solid (injection). The valence band is modified similarly to the
conduction band by the metallic contact and the external field. The process is specular,
so that injection and extraction of holes take place at x = 0 and x = L, respectively.
ϕ=0 ϕ=0
Ec
_ +
Electrics field due to +
_
_ _
_
_
+
+
depletion regions and +
+
_
_
surface charges
+ +
_ + +
_

x=0 x=L
ϕ>0 ϕ=0
Ec
_ _
External electric field
+ +

x=0 x=L
Figure 4.7: Injection and extraction. Blue areas represent the metal, grey ones the non-metallic solid.
(Above) Without an external field a depletion region grows at the interfaces, raising the conduction
band. (Below) When the field is applied the conduction band (blue curve) is raised at x = 0 and
lowered at x = L, allowing for extraction and injection of carriers. The depletion region is still
present, but it has been omitted for the sake of clarity.

48
 4.2. Band Theory of Solids: Key Concepts

4.2.7 Drift and Diffusion

Consider a non-metallic solid subject to an external electric field, E(t). Under its driv-
ing force, the random motion of free carriers is perturbed in such a way that an average
non-zero velocity arises in the field direction, giving origin to the drift current. In
semiconductors the drift current is influenced by the collisions between charge carri-
ers and the lattice, that yields a chaotic motion with a mean shift in the field direction
(Figure 4.8, left). Insulating materials, on the other hand, do not have an extended crys-
talline structure, but they rather contain amorphous and semi-crystalline regions that
significantly reduce the mean distance covered by carriers. Jonscher suggested that the
dominant drift mechanism in insulators may be the hopping of charge carriers in shal-
low traps [14] (Figure 4.8, right). The drift current can be expressed in two equivalent
ways,
   
Jdr = −e vdr,n T, |E(t)|, t n T, |E(t)|, t
   
+ e vdr,p T, |E(t)|, t p T, |E(t)|, t
 
= σ T, |E(t)|, t E(t), (4.15)

where the expressions stress the dependence of drift velocities of electrons and holes,
vdr,n and vdr,p respectively, carriers’ densities, n and p, and conductivity, σ, on tem-
perature, electric field intensity and time. The behaviour of σ is determined by the
dominant mechanisms that rule the drift process.

E
Ec
_

Field direction
+

Ev

Figure 4.8: (Left) Chaotic motion of charge carriers due to the collision with the lattice. There is an
average motion in the direction of the field. (Right) Hopping of charge carriers in shallow traps close
to the conduction and valence bands.

The free carriers’ densities, n and p, are not necessarily uniform in the solid, and
when this happens another type of current, the diffusion current, arises as a consequence
of the concentration gradients:

Jdf = e(Dn ∇n − Dp ∇p), (4.16)

where Dn and Dp are the diffusion coefficients for electrons and holes, respectively.
While the drift current reacts immediately to the application of an external electric
field, the diffusion current requires the build up of gradients ∇n and ∇p, giving rise to
a much slower dynamics, as observed by Jonscher [14].

49
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

4.3 Fundamental Assumptions, Modelling of Deep and Shallow Traps

The rest of this chapter is dedicated to the derivation and validation of a model that
describes the dynamic behaviour of polymeric insulating materials under suitable sim-
plifying assumptions. This work has been published in [18]. The proposed model relies
on the work from Shockley and Read [19] on the statistics of recombination by trap-
ping and detrapping processes, that led to the theory of thermally stimulated currents
developed by Simmons, Taylor and Tam, which is used to identify the distribution of
trapping states in insulators [20–22]. The simplifying assumptions and their implica-
tions will now be stated. To fix the ideas, it is useful to recall the configuration of
Figure 2.1 in which materials are tested.
(I) Low electric field. The electric field is relatively low, up to 30 kV/mm. This field
is close to those used for DC cables’ design, and there seems to be evidence that
in this range the dynamic is dominated by trapping and detrapping processes [39].
Recombination and generation, therefore, can be neglected.
(II) Single charge carrier. Under low electric field and negligible recombination there
is some evidence, at least in polyethylene-based polymers, that the net trapped
charge is unipolar [40, 41]. The model considers one charge carrier.
(III) Prevailing hopping conduction. Diffusion of charge carriers is neglected, which
is a common assumption [?, 39, 40]. The transport through the bulk occurs mainly
via hopping in shallow traps [?, ?, 14].
(IV) Infinite availability of carriers. The flow of carriers injected and extracted at the
metallic contacts is limited by the number of energy levels involved in these pro-
cesses. The physical laws for injection and extraction are not trivial, and their
parameters cannot be inferred from the PDC measurements on which the model
is validated. Therefore, it is assumed that the number of states is sufficiently large
that the number of carriers that can potentially flow into the solid is infinite and
no specific injection nor extraction law is necessary.
These assumptions can be, partially, sketched by the energy band diagram in Figure 4.9.
Deep traps are involved in the capture and release of carriers, while hopping guides
them through shallow traps. The two types of traps exchange carriers only with the
conduction band, not with each other. Both the Fermi level and the valence band are at
−∞ on the energy axis. The assumption of non-interacting traps requires two separate
models for each of them, that will be developed in the next subsections.

4.3.1 Deep Traps

A model for deep traps can be derived following the approach of Shockley and Read
[19]. Consider a unit volume of the material. There will be a total number
Nc (E) dE (4.17)
of energy levels in the range dE of the conduction band. Being f (E) the Fermi-Dirac
statistics of the carriers, the number of occupied levels will be
f (E)Nc (E) dE. (4.18)

50
 4.3. Fundamental Assumptions, Modelling of Deep and Shallow Traps

E
Ec
Hopping

Emission
Capture
Shallow traps

Deep traps

Figure 4.9: Energy band diagram of the polymer model after the simplifying assumptions. Deep traps
are involved in trapping (capture) and detrapping (emission) of carriers. Shallow traps take part in
the hopping process. The valence band, and the Fermi level, are not relevant to the model.

The occupied levels host free carriers that can be captured by the deep traps. The
probability of capture is measured by the cross-section of the deep traps, ct . The cross-
section is in general energy-dependent, but it will be assumed constant in the proposed
model. Multiplying ct by the average drift velocity of carriers in the conduction band,
vdr , yields the probability per unit of time that a carriers is captured by a deep trap,
hence the fraction of carriers captured per unit of time is

vdr ct f (E) Nc (E) dE, (4.19)

under the assumption that all the traps are empty. Of course, only free traps can partic-
ipate to the capturing process. According to Shockley and Read [19], the occupation
probability of a deep trap can be described by a Fermi-Dirac statistics, ft (Et ), where
Et is the energy of the deep trap in the forbidden gap. Therefore, the probability of
finding an empty trap is 1 − ft (E), and the effective number of captured electrons is

[1 − ft (Et )] vdr ct f (E) Nc (E) dE. (4.20)

In this expression Nc (E) is measured in J−1 m−3 , ct in m2 , v in ms−1 , dE in J and the

remaining quantities are dimensionless, hence the entire expression has the dimension
of s−1 that can be interpreted as a trapping rate. The total capturing rate, rc , is obtained
by integrating (4.20) over the conduction band energy, dE:
Z +∞
rc (Et ) = [1 − ft (Et )] vdr ct f (E) Nc (E) dE. (4.21)
−∞

Bringing outside of the integral the quantities independent from the integration variable,
E, yields
Z +∞
rc (Et ) = [1 − ft (Et )] vdr ct f (E)Nc (E) dE
−∞

= [1 − ft (Et )] vdr ct n, (4.22)

where n is the density of carriers in the conduction band and it is given by (4.5). The
capturing rate is useful to calculate the total charge density captured by the deep traps

51
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

in the unit of time, ḃc , obtained by the following integration:

Z +∞
ḃc = e rc (Et ) g(Et ) dEt , (4.23)
−∞

where g(Et ) is the DOS in the forbidden gap and e > 0 is the electron elementary
charge. An observation is necessary here. The upper bound of the integral in (4.23)
is +∞, but it should be, strictly speaking, Ec . On the other hand, the energy levels
allowed by g(Et ) lie, by all practical means, in the forbidden gap, therefore the integral
can be extended to +∞ without mistakes. The total charge density stored by deep traps
is given by Z +∞
b=e ft (Et ) g(Et ) dEt , (4.24)
−∞
and the total charge density that can be theoretically stored is
Z +∞
N =e g(Et ) dEt . (4.25)
−∞

The drift current, due only to carriers in the conduction band, is given by
Jdr = e vdr n. (4.26)
With these definitions (4.23) becomes
Z +∞
ḃc = e vdr ct n [1 − ft (Et )] g(Et ) dEt
−∞
" Z #
+∞ Z +∞
ct Jdr
= e g(Et ) dEt − e ft (Et ) g(Et ) dEt
e −∞ −∞

N −b
= ct Jdr . (4.27)
e
It is useful, at this point, to make the following consideration. In the event of a
capture, a carrier falls, most likely, to the lowest available non-occupied trap. In the
event of an emission, the released carrier comes, most likely, from the highest occupied
trap. These are consequences of the Fermi-Dirac statistic. By all practical means the
probabilistic treatment can be removed from this situation, which means approximating
the Fermi-Dirac statistics with a step function. With this logic the trapped charge, b,
becomes a function of the highest occupied level, Em , and it is computed as
Z Em
b(Em ) = e g(Et ) dE, (4.28)
−∞

and it is no longer necessary to employ ft as in (4.24), because the calculation assumes

that all the energy levels below Em are occupied. Consequently, the rate of capture of
the deep traps is given by
N − b(Em )
ḃc (Em ) = ct Jdr . (4.29)
e
52
 4.3. Fundamental Assumptions, Modelling of Deep and Shallow Traps

Consider now the emission of a carrier from a deep trap. According to Shockley and
Read, [19], the probability of emission per unit of time is
pe = vdr ct e(Et −E)/kB T , (4.30)
where E is in the conduction band and Et is in the forbidden gap. Hence, the emission
rate depends on the distance, Et − E, between the emitting and receiving energy levels.
Recalling the discussion of section 4.2 and in particular equation (4.6), the conduction
band was assumed to be concentrated in a single energy level, Ec , therefore E = Ec .
Since energies are defined up to an additive constant, it is possible to let Ec = 0, so that
the emission probability becomes
pe = vdr ct eEt /kB T . (4.31)
Since the first emitted electron comes form the trap at the highest level, it follows also
that Et = Em , yielding
pe (Em ) = vdr ct eEm /kB T . (4.32)
Following again the general approach of Schokley and Read, the number of electrons
emitted to the conduction band depends on the number of available state, i.e.
[1 − f (E)] Nc (E) dE, (4.33)
but consider that, now, Nc (E) = Nc δ(E). Thus, the total emission rate in the conduc-
tion band, re , is given by
Z +∞
re (Em ) = pe (Em ) [1 − fe (E)] Nc (E) dE. (4.34)
−∞
Developing the calculations in (4.34) yields
Z +∞
re (Em ) = pe (Em ) [1 − f (E)] Nc (E) dE
−∞
"Z #
+∞ Z +∞
= pe (Em ) Nc (E) dE − f (E)Nc (E) dE,
−∞ −∞

"Z #
+∞ Z +∞
= pe (Em ) Nc δ(E) dE − f (E)Nc (E) dE,
−∞ −∞

= pe (Em )(Nc − n). (4.35)

A consequence of the hypothesis of infinite availability of charge is that the number
of available states in the conduction band is much larger than the effective number of
carriers, i.e.
n ≪ Nc , (4.36)
and the emission rate can be written as
re (Em ) ≃ pe (Em )Nc

= vdr ct eEm /kB T Nc

= ν0 eEm /kB T , (4.37)

53
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

with ν0 = vdr ct Nc being the attempt-to-escape frequency. The total charge density
emitted from the deep traps is then computed by integrating over the filled traps:
Z Em
ḃe = e re (Em ) g(Et ) dEt
−∞
Z Em
Em /kB T
= ν0 e e g(Et ) dEt
−∞

= ν0 eEm /kB T b(Em ). (4.38)

4.3.2 Shallow Traps

Modelling the shallow traps is a more difficult task. Their proximity with the conduc-
tion band does not allow them to be treated as independent trapping states. A possible
solution that avoids excessively complicated models is to pursue a heuristic approach,
by observing the transient behaviour of polymeric materials and inferring an empirical
law from it. Consider a PDC measurement at fixed temperature and uniform field. It
will be shown that a time-dependent conductivity, σ(t), linking the drift current to the
applied constant electric field, E(t), offers a rather satisfactory compromise between
simplicity of the model and accuracy in the reproduction of the experimental results:
Jdr (t) = σ(t)E(t). (4.39)
The empirical time-dependent conductivity holds all the processes contributing to the
drift current, which, for the present model, are the average motion of electrons in the
conduction band and their hopping in shallow traps.

4.4 The Journey of an Electron

The previous section introduced the foundational hypotheses of the model, derived the
simplified equations for the dynamic of deep traps, and proposed a general empirical
model for the drift current. The discussion also clarified the energy band structure
assumed by the model, shown in Figure 4.9, which separates deep and shallow traps,
and deems them as non-interacting. This picture, however, is not complete. Electrons,
the only carrier considered after hypothesis (II), have a double dynamic: one among the
energy bands, and the other one in space. This section discusses the second dynamic.
First of all, consider the scheme in Figure 4.10. Hypotheses (I), (III) and (IV) imply
that recombination, generation and diffusion of carriers, and limits to injection and ex-
traction of carriers, are all neglected by the model. Since these phenomena determine
the space charge dynamic, the immediate consequence is that the proposed model ne-
glects the dynamic of space charge. Let us clarify immediately that by space charge it
is meant the free charge density, ρ, that is given by
ρ = ∇ · D, (4.40)
where D is the displacement field in the polymer. With this premise in mind, the
“journey” of the spatial dynamic of the electron follows from a thorough discussion
of Figure 4.11. The discussion assumes that a flat sample of the polymer is placed

54
 4.4. The Journey of an Electron

(I) Low electric (III) Prevailing hopping (IV) Infinite availability

(I) field (III) conduction (IV) of carriers

Neglects

Neglects

Neglects
Recombination Limits to injection and
Diffusion
and generation extraction of carriers

Phenomena determining
space charge dynamic

Figure 4.10: Hypotheses (I), (III) and (IV) impose that the space charge dynamic is negligible in the
proposed model.

between two metallic plates: a ground (GND) electrode, which is connected to ground
through an ammeter, and a high voltage (HV) electrode, which is connected to a voltage
source, vs (t) (see Figure 4.12 for a detailed picture). An electron is injected from the
GND electrode into the polymer, overcoming the potential barrier. After being injected,
the electron immediately finds itself in the conduction band of the polymer, initially
contributing to the drift current, Jdr . At this point, there are three possible evolutions
of the situation.
1. The electron continues moving through the bulk of the polymer, drifting in the
conduction band and occasionally hopping in shallow traps, then reaches the in-
terface between the polymer and the HV electrode, and jumps back into the metal.
2. The electron is immediately trapped by the surface states at the interface between
the polymer and the GND electron, forming a layer of surface charge density.
3. The electron moves through the bulk of the polymer and at some point is captured
by a deep trap in the bulk, giving rise to a space charge density, ρ. At this point
the electron continues to move through the bulk, but at a much slower pace, taking
part in the space charge dynamic.
The proposed model neglects the space charge dynamic. However, the ammeter in
Figure 4.12 will measure a current, im (t), that is inevitably affected by space charge. In

Injected electron

Initially contributes to Jdr

in the conduction band

Continues drifting Immediately trapped Trapped at some

and hopping by surface states point in the bulk Space charge
dynamic
Proposed model

Moves slowly
in the bulk

Figure 4.11: The journey of an electron after being injected into the conduction band of polymer, with
its possible pathways.

55
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

vs(t)
Induced surface
HV electrode + + + + + +
charges

Polymeric Space

Drift
E(t) _ Ddt(t)
sample charge
_ _ _ _ _ _
_
GND electrode Trapped (captured)
im(t) surface electrons

A Ammeter

Figure 4.12: Simple scheme of the experimental setup used to measure PDC in polymers. Source voltage
vs (t) is applied to the HV electrode, while the GND is connected to ground through an ammeter.
Electrons are indicated by the blue, negative charges, while the red charges are positive (they are not
holes). E(t) is the electric field associated to the voltage gradient in the polymer, and Ddt (t) is the
displacement field stemming from the trapped electrons and the induced surface charges. An electron
drifting in the bulk and another one taking part in the space charge dynamic are also shown. Current
im (t) is the one measured by the ammeter.

vs(t)

0 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

q0
ρ D0 q0 (ρ,x) + qd (ρ,x)
x _____________
_ _ __ _ _ _
_ _ _
_ _
_ _ _ _ __ _ _
_ _ _ _ _
_ _ _ _
Db(ρ,x) =
Dd qd A
d _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

im(t)

A Ammeter

Figure 4.13: A thin layer of charge density ρ is located at distance x from the HV electrode and at
distance d − x from the GND electrode. The thickness of the sample is d. Charges q0 and qd appear
on the surfaces of the sample as a consequence of Gauss’ law. The displacement field is split into D0
and Dd , but an equivalent field Db can be defined to describe the entire situation.

the interpretation of the experimental and theoretical results, the philosophy will be to
treat the model predictions as the ideal behaviour of the material, while the space charge
dynamic will be a disturbance. The rationale behind this philosophy originates from an
understanding of the logic that allows the ammeter to detect the transient phenomena
occurring in the polymer (Figure 4.12). Two scenarios can be distinguished: trapping of
an electron at the interface between GND electrode and polymer; motion of an electron
inside the bulk of the material.
Trapping at the interface When an electron is trapped at the interface between the
GND electrode and the polymer, it contributes to the formation of a surface layer
with negative charge (Figure 4.12). It can be speculated that this may be possible
thanks to the surface trapping levels mentioned in section 4.2.3, that are deep
traps at the interface. In light of Gauss’ law, the total surface charge must be
balanced by a space charge in the polymer. Since no charge is introduced into the

56
 4.4. The Journey of an Electron

bulk by this process the total surface charge must be zero, thus a positive charge
of equal intensity arises on the other interface, between the polymer and the HV
electrode (Figure 4.12). The positive and negative surface charges are associated
to a displacement field, Ddt , which is solenoidal, ∇ · Ddt = 0, so as not to give
rise to any charge distribution in the bulk. The deep surface traps contribute to
the measured current, im (t), by means of the time derivative of the displacement
field, Ḋdt .
Motion inside the bulk When an electron is injected into the bulk of the polymer, it
contributes to the build up of a space charge density, ρ. For the sake of simplic-
ity, the space charge will be assumed to be distributed in a thin region placed at
distance x from the HV electrode, and at distance d − x from the GND electrode,
being d the thickness of the sample, as shown in Figure 4.13. The negative charge
layer divides the polymer’s volume into two regions, each of which has a given
displacement field: D0 (ρ, x) in the upper region, and Dd (ρ, x) in the lower region
(see Figure 4.13). Both fields depend in general on the bulk charge density, ρ, and
on its position in the polymer, x. The total charge accumulated at the interfaces
between the polymer and the electrodes is
q0 (ρ, x) + qd (ρ, x) = −D0 (ρ, x)A − Dd (ρ, x)A, (4.41)
where A is the area of the sample. To simplify the following discussion, it is
possible to define an equivalent displacement field representative of the bulk,
q0 (ρ, x) + qd (ρ, x)
Db (ρ, x) = . (4.42)
A
This field contributes to the measured current by means of its time derivative,
∂Db ∂Db
Ḋb = ρ̇ + ẋ. (4.43)
∂ρ ∂x
If the charge density does not change significantly, ρ̇ ≃ 0, then ∂x Db ẋ describes a
constant charge moving through the bulk in the x-direction. In other words, it is a
drift current,
∂Db
Ḋb ≃ ẋ = Jdr . (4.44)
∂x
If, instead, the charge moves very slowly, so that ẋ ≃ 0, the current
∂Db
Ḋb ≃ ρ̇ = Ḋsc (4.45)
∂ρ
describes an accumulation or a shrinkage of charge density, which is the dynamic
of space charge in a more strict sense.
The two scenarios follow a very simple logic: whatever happens inside the bulk or on
the surface of the polymer, is reflected in a variation of the surface charge at the in-
terfaces between the polymer and the electrodes. The variation of the surface charge
induced on the electrodes’ faces will cause the circulation of a current in the measure-
ment circuit, im (t), which is detected by the ammeter. In mathematical terms, assuming
that all the vector fields are directed along the x-axis, as in Figure 4.13 and Figure 4.12,

57
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

the measured current is proportional to the time derivative of the total displacement
field D:
im (t) ∝ Ḋ = Ḋdt + Ḋb = Ḋdt + Jdr + Ḋsc , (4.46)
which takes into account the dynamic of the charge captured by the deep surface traps,
Ḋdt , the drift current, Jdr , and the space charge, Ḋsc . Recalling Figure 4.11, the pro-
posed model considers only the trapping by surface states and the drift current, and the
associated current is  
imod (t) = A Ḋdt + Jdr , (4.47)

while the current due to the space charge is

isc (t) = AḊsc . (4.48)

Therefore, the measured current is

im (t) = imod (t) + isc (t), (4.49)

which is the mathematical way of saying that space charge, isc (t), is a disturbance on
the ideal behaviour of the material, imod (t). Equation (4.49) can be used to calculate
the space charge stored in the sample,
Z +∞ Z +∞
 
Qsc = isc (t) dt = im (t) − imod (t) dt. (4.50)
0 0

Equation (4.50) should not be considered as a precise calculation of the stored space
charge. First of all, current im (t) is affected by measurement errors and it cannot be
measured for infinitely long times. Second, recalling the discussion at the beginning of
this chapter, the theory behind imod (t) deems charge carriers as non-interacting parti-
cles. This is a strong simplification, but also the very reason that allowed the separation
of surface traps, drift current and space charge contributions in (4.46). Equation (4.50),
then, should be used as a mean to provide an estimate of the stored space charge.

4.5 The Equivalent Circuit

The model under development considers three different contributions to the measured
current, im (t): the dynamic of deep traps, the dynamic of the drift current and, in
addition, the contribution of fast and, potentially, slow polarization processes described
by the Debye model. Considering an applied uniform electric field, E(t), as the one in
Figure 4.12, the measured current takes the following form:
 
XM
im (t) = A ε∞ Ė(t) + Ṗk (t) + Jdr (t) + Ḋdt (t) . (4.51)
k=1

The first term is the current associated to the fast polarization processes,
ε∞ A
if p (t) = Aε∞ Ė(t) = · dĖ(t) = C∞ v̇s (t), (4.52)
d
58
 4.5. The Equivalent Circuit

where vs (t) is the voltage applied to the HV electrode in Figure 4.12. The second term
is the current associated to the slow polarization processes,
M
X
isp (t) = AṖk . (4.53)
k=1

Without loss of generality, in the following only M = 1 Debye relaxation process will
be considered. Recalling the expression of the drift current density given in (4.39), the
third term in (4.51) becomes
Aσ(t)
idr (t) = Aσ(t)E(t) = · dE(t) = G(t) · vs (t). (4.54)
d
The last term models the dynamic of the deep traps,
idt (t) = AḊdt (t) = Adḃ, (4.55)
where b is the trapped charge density, given by (4.28), and ḃ is its time derivative. In
(4.55) ḃ has been multiplied by the sample volume, Ad, because b was considered as
a volume density in section 4.3.1. The discussion of the previous section, however,
stated that deep-trapped electrons considered by the model were on the surface, hence
b should be a surface charge density, strictly speaking. The proposed model however
is macroscopic, and even if it was stated that it considers only the electrons trapped on
the surface, there is no reason to exclude a possible influence of the bulk deep traps
immediately adjacent to the surface. The matter is more of an interpretation issue,
that cannot be resolved using the simplified approach discussed here. Therefore, for
the sake of generality, b will be considered as a volume charge density and expression
(4.55) will be the governing one. The net accumulation of charge, ḃ, comes from a
balance between capture and escape, ḃc − ḃe . Hence, using (4.29) and (4.38),
idt (t) = Adḃc − Adḃe
N −b
= Adct Jdr (t) − Adν0 eEm /kB T b
ct  e 
= Jdr (t) N e − B − ν0 eEm /kB T B, (4.56)
e
where Ne = AdN and B = Adb. The first term in (4.56) is the capture current,
ct    
ic (t) = · AJdr (t) · N − B = S · idr (t) · N − B ,
e e e (4.57)
Ae
with Se = ct /Ae. In the second term charge B is multiplied by a quantity whose
dimension is the inverse of a second (ν0 is the attempt-to-escape frequency), hence it
can be interpreted as the inverse of a time constant:
τ (Em ) = ν0−1 e−Em /kB T = τ0 e−Em /kB T , (4.58)
with τ0 = ν0−1 . The expression of τ is written in terms of the energy of the highest
occupied trap, Em , but observing that the relationship
Z Em
B(Em ) = Ad g(E) dE (4.59)
−∞

59
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

is invertible (g(E) is non-negative, thus the integral is a monotonic function of Em ), it

is possible to reverse B(Em ) and write Em (B), thus
τ (B) = τ0 e−Em (B)/kB T (4.60)
and the escape current can be written as
B
ie (t) = . (4.61)
τ (B)
The current associated to the deep traps, in conclusion, is given by:
idt (t) = Ḃ = ic (t) − ie (t)
  B
= S idr (t) N − B −
e e . (4.62)
τ (B)
At this point it is possible to write im (t) in its final form:
im (t) = if p (t) + isp (t) + idr (t) + idt (t)
= C∞ v̇s (t) + isp (t) + G(t) vs (t) + Ḃ. (4.63)
Equation (4.63) corresponds to the equivalent circuit in Figure 4.14. The branches
associated to currents G(t)vs (t), C∞ v̇s (t) and isp (t) are immediately recognized, as far
as one recalls that the slow polarization current isp (t) is described with a series R-C
branch. The last branch on the right corresponds to the deep traps’ current, idr (t) = Ḃ.
In particular, equation (4.62) is a Kirchhoff’s current law at node a in Figure 4.14.
Current Ḃ charges a non-linear capacitor described by a charge-dependent elastance
Sdt (B). The elastance is related to the time constant τ (B) by
1
τ (B) = , (4.64)
Gdt Sdt (B)
Gdt being a conductance. Indeed, consider the circuit of Figure 4.14 when no voltage
is applied, namely vs (t) = 0. Without an applied voltage there is no drift current,
idr (t) = 0, and therefore no capture, ic (t) = 0. Under these conditions (4.62) reads
B Ḃ
Ḃ + = Ḃ + Gdt Sdt (B) B = + Sdt B = 0, (4.65)
τ (B) Gdt
which is the Kirchoff voltage law for the circuit in Figure 4.15.

4.6 Experimental Facts

The central idea of the model, culminated in the circuit of Figure 4.14, is the inde-
pendence of physical processes, and it has two interesting consequences. The first one
is the modularity: if a new processes is found its corresponding equation is added to
the model and the circuit is extended with a new branch. Second, each process has a
characteristic timespan: some of them expire in a few seconds, others require hours.
Thus, at different stages of the transient only some of the branches in Figure 4.14 will
actually contribute to the measured current, im (t). Identifying the dominant process at
each stage is extremely helpful in fitting the model parameters to the experimental data.

60
 4.6. Experimental Facts

vs(t)

ic(t) ie(t)

G(t) C∞ a
.
. B
G(t)vs(t) C∞ vs(t) isp(t) Sdt(B)

im(t)
A

Figure 4.14: The equivalent circuit of the polymer model. Each parallel branch provides the current
originating from a specific physical process, that contributes to the total current im (t) measured by
the ammeter.

Gdt
. a
B
Sdt(B)

Figure 4.15: Representation of the branch associated to the deep traps under short-circuit conditions,
vs (t) = 0.

To validate the model, three polypropylene (PP) samples and one cross-linked polyethy-
lene (XLPE) sample were tested. The experiments were performed at the Laboratory
of Innovative Materials for Electrical Systems (LIMES) of the University of Bologna,
using a typical cell for PDC measurements [17]. Table 4.1 summarizes the testing
conditions and the geometric characteristics of each sample.
Table 4.1: Testing field and temperature, and geometric characteristics of the polymeric samples.

Sample Electric field (average) Temperature Thickness Area

PP1 10 kV/mm 30 ◦ C 0.5 mm 5.31 cm2
PP2 30 kV/mm 30 ◦ C 0.5 mm 5.31 cm2
PP3 30 kV/mm 30 ◦ C, 60 ◦ C, 90 ◦ C 0.35 mm 5.31 cm2
XLPE 30 kV/mm 20 ◦ C, 60 ◦ C, 90 ◦ C 0.5 mm 5.31 cm2

Figure 4.16 shows the PDC test performed on PP1. The sample underwent a se-
quence of five polarization and depolarization cycles. The polarization time was in-
creased from 10 s to 3 · 105 s. The depolarization time was fixed at 105 s. When the volt-
age was applied for the first time, polarization current A was measured (Figure 4.16, left
plot). Then the voltage was removed, and the ammeter measured depolarization current
B (Figure 4.16, right plot). Then the voltage was reapplied and current C was observed,
and so on for the remaining cycles. The result of Figure 4.16 shows an evident fact:
the duration of the polarization transient does not affect the mean trend of
the subsequent depolarization transient.

61
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

Polarization Depolarization

Figure 4.16: Polarization and depolarization currents measured in PP1. The test consisted in five full
polarization and depolarization cycles, with increasing charging time and fixed discharging time.
The alphabetic order of the labels follows the chronological order of the transients.

During depolarization the electrodes are short-circuited, since vs (t) = 0. Therefore,

the branches of the drift current, idr (t), and of the fast polarization current, if p (t), do
not contribute to the measured current, so, in principle,
im (t) = isp (t) + Ḃ during depolarization. (4.66)
The discussion of chapter 3, however, showed that the R-C branch is not a proper
model for the discharge current in polymers, hence the conclusion is that
im (t) = Ḃ during depolarization. (4.67)
In other words, the depolarization current is due to the emptying of the deep traps. The
discharge transient is very long and its trend is independent form the charging duration,
the logical conclusion then is that
the deep traps are immediately charged as voltage is turned on, then they are
released very slowly.
This asymmetric behaviour is typical of deep traps, which can easily capture the charge
carriers, but are much less prone to release them. This fact has another important
implication. During polarization, current Ḃ will fall to zero within the first 10 s (the
duration of current A in Figure 4.16). With a fixed DC voltage, vs (t) = V , also current
if p (t) will be zero, as suggested by equation (4.52). Neglecting again the contribution
of the R-C branch, that should be marginal, the measured current during polarization
will coincide with the drift current:
im (t) = idr (t) = G(t)V during polarization. (4.68)
The polarization curves in Figure 4.16 resemble a power law approaching the steady
state, and equation (4.68) indicates that the transient is associated to the time varying
conductivity G(t). This is the experimental fact that confirms the empirical law that
was suggested for the drift current in equation (4.39). A suitable expression for the
conductance during polarization is
G(t) = G2 + (G1 − G2 ) (1 + t)−n , (4.69)

62
 4.6. Experimental Facts

Polarization Depolarization

Figure 4.17: Polarization and depolarization currents measured in PP2. The test consisted in four full
polarization and depolarization cycles, with fixed charging time and decreasing discharging time,
plus one final polarization transient. The alphabetic order of the labels follows the chronological
order of the transients.

where G1 is the conductance at 0 s, G2 is the steady-state value and n is a suitable fitting

coefficient.
Figure 4.17 shows the PDC test performed on PP2. The sample underwent a se-
quence of four polarization and depolarization cycles, plus one final polarization tran-
sient. The polarization time was fixed at 105 s, while the depolarization time was de-
creased from 104 s to 5 s. The interpretation of the currents’ sequence is the same of
Figure 4.16. It can be observed that the depolarization currents in Figure 4.17 follow
the same trend in each cycle, but the same does not happen for the polarization ones.
Let us focus on the highlighted region in the plot of the polarization currents, which
covers the time range between 20 s and 40 s. This range coincides with the first stage
in which the generator is at steady state, vs (t) = V . Indeed, the voltage source used in
the experiments had an intrinsic exponential transient,
 
−t/τg
vs (t) = V 1 − e during polarization, (4.70)

with τg = 2.5 s. The same transient occurred when voltage was turned off,

vs (t) = V e−t/τg during depolarization. (4.71)

In the highlighted region currents A and C are similar. Since the depolarization transient
between them is relatively long (104 s, current B), a long depolarization transient does
not seem to affect the subsequent polarization. If the depolarization transient duration
is halved to 5 · 103 s, as for current D, the matter is different. Looking at the sequence
of transients C - D - E it is evident that halving the depolarization transient causes
a reduction of current E with respect to C, in the highlighted region. This suggests
that shorter depolarizations have an impact on the subsequent polarizations. The same
argument applies to the sequence E - F - G, in which the depolarization current F lasts
only for 50 s. An extreme case is that of current H, since its transient holds only for 3 s:
here the difference between the previous and the subsequent polarizations, currents G
and I, is negligible. Summarizing, the general principle is that

63
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

G1

G2

Figure 4.18: Time-varying conductance. During polarization G(t) is a power law (solid, black line). If
the polarization is sufficiently long, G(t) reaches the steady-state value, G2 (dashed, black line). If
the voltage is removed and the depolarization transient begins, the conductance rises to the initial
value G1 , following an exponential law (red line).

the duration of the depolarization transient affects the subsequent polariza-

tion transient.
Since the measured current during polarization is dominated by the drift current, as
highlighted by equation (4.68), the variable duration of the depolarization transient
should be reflected in the model of the time-varying conductivity, G(t). The idea,
explain with the help of Figure 4.18, is the following. During polarization, the conduc-
tance falls from the initial value G1 to the steady-state value G2 . During depolarization,
a conductance-restoring process brings G(t) back to the initial value G1 by means of
an exponential transient with time constant τa . If the depolarization transient is stopped
before the restoring process has been completed, the subsequent polarization will start
with a conductivity lower than G1 , thus the polarization current will start from a lower
value. The model of G(t) can be generalized as
 
G(t) = a(t) G1 + 1 − a(t) G2 , (4.72)
where a(t) is governed by the following differential equations:
( n+1
ȧ + na n = 0 if vs (t) ≥ Vtr
(4.73)
ȧ + a−1
τa
=0 if vs (t) < Vtr

where Vtr threshold voltage and the initial condition is a(0) = 1. The first equation
in (4.73) models the power law, while the second equation models the conductance-
restoring process.

4.7 Density of States of the Deep Traps

The development of the model is missing one last element, before it can simulate the
behaviour of polymeric materials. The escape current of the deep traps was written, in

64
 4.7. Density of States of the Deep Traps

section 4.5, as
B
ie (t) = , (4.74)
τ (B)
with
τ (B) = τ0 e−Em (B)/kB T , (4.75)
and the relationship between B and Em was
Z Em
B(Em ) = Ad g(E) dE. (4.76)
−∞

The missing element is the definition of the DOS, g(E). Experiments based on surface
potential decay and thermally stimulated depolarization currents suggest that the trap
distribution is a superposition of Gaussian distributions [?, 22, 42], but in the view of
simplifying the approach, the proposed model will assume a single Gaussian for the
DOS. The expression of g(E) then is
" #
N (∆E + E)2
g(E) = √ exp − , (4.77)
α 2π 2α2

where N is the theoretical maximum charge density that can be captured by the traps,
and it is measured in C/m3 , while the meaning of ∆E and α is indicated in Figure 4.19.
Recalling that Ne = AdN , the relationship between B and Em can now be written
explicitly as " #
Ne Z Em (∆E + E)2
B(Em ) = √ exp − dE, (4.78)
α 2π −∞ 2α2
while the inverse relation, appearing in (4.75), is
√
 
2B
Em (B) = −∆E + 2α erf−1 −1 , (4.79)
Ne

being erf−1 the inverse error function. At this point it is convenient to do a slight
modification to the expression of τ (B). With the expression (4.75), τ = τ0 when
Em = 0, which is of little physical significance. If the time constant is rewritten as
 
∆E + Eav (B)
τ (B) = τ0 exp − , (4.80)
kB T

2α

g(E)

Eav -ΔE Em E
Figure 4.19: Gaussian DOS of the deep traps. The distribution centre is ∆E, while α is the term usually
referred as standard deviation. The blue area represents the portion of filled traps.

65
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

where Eav is the average occupation energy, then the value τ0 is reached when Eav =
−∆E, that is when all the traps are filled: this can be deemed as a new definition of τ0 .
The average occupation energy is calculated as
Z Em
Eg(E) dE
−∞
Eav (Em ) = Z Em
g(E) dE
−∞
 2
∆E+E m
− √
r
2 e 2α
= −∆E + α   , (4.81)
π erfc ∆E+E
√ m −2
2α

where erfc is the complementary error function, and using (4.79) yields
r
2 N e −herf−1 (2B/Ne −1)i2
Em (B) = −∆E − α e . (4.82)
π 2B
In conclusion, the final expression for the time constant is
(r )
2 α N e −herf−1 (2B/Ne −1)i2
τ (B) = τ0 exp e . (4.83)
π kB T 2B

Note that the trap distribution centre, ∆E, does not appear in (4.83), hence it does not
have to be estimated.

4.8 Fitting of PDC Measurements

The theoretical presentation of the model, along with the inference of its main prop-
erties, is now complete. The remaining task is to verify its capability in reproducing
the experimental tests performed on samples PP1, PP2, PP3 and XLPE. In order to
reproduce the experiments, it is first of all necessary to identify the number of free
parameters. The measured current has five components, namely

im (t) = idr (t) + if p (t) + isp (t) + ic (t) − ie (t), (4.84)

whose expressions were derived in the previous sections. Table 4.2 recalls the general
expression of each components and lists the free parameters in each of them. The
free parameters can be referred either to the circuital model, or to the physical model.
Conductivities σ1 and σ2 were not mentioned in the previous sections, but they can
easily be derived from G1 and G2 by means of the scaling factor d/A:
d d
σ1 = G1 and σ2 = G2 . (4.85)
A A
The general expression of isp (t) was not given in the table because it depends on the
applied voltage waveform. In any case, isp (t) represents a Debye relaxation process,
thus its parameters are either the time constant and the capacitance of the R-C branch,
or the time constant and the relative permittivity of the polarization process.

66
 4.8. Fitting of PDC Measurements

Table 4.2: List of the free parameters involved in the model.

Free parameters
Term General expression
Circuital model Physical model
idr (t) G(t)vs (t) G1 , G2 , n, τa , Vtr σ1 , σ2 , n, τa , Vtr
if p (t) C∞ v̇s (t) C∞ ε∞
isp (t) – τsp , Csp τsp , εsp
ic (t) Se idr (t) (N
e − B) S,
e N e ct , N
B
ie (t) τ (B) τ0 , α, N
e τ0 , α, N

The total number of free parameters listed in Table 4.2 is 12. However, the ex-
perience gained in fitting the experiments allowed to reduce this number. The time
constant τ0 , the traps’ cross section, ct , and the threshold voltage Vtr of the conduc-
tance model, do not change appreciably from one material to another: τ0 was fixed at
10−12 s, which is the reciprocal of a typical value for the attempt-to-escape frequency,
1012 Hz [43–45]; the traps’ cross-section was empirically estimated to 10−14 m2 ; the
threshold voltage was fixed at 100 V. The number of parameters is then reduced to 9.
Moreover, one must recall that the slow polarization currents may not be necessarily
detectable in polymers, since they might be dominated by the other processes. There-
fore, if isp (t) does not provide any significant contribution, the number of parameters
can be further reduced to 7.

4.8.1 PP1 and PP2 Transient Reconstruction

Figure 4.20 shows the fitting of the PDC in PP1. In both figures only one fitting curve
is shown, since the behaviour of both the polarization and depolarization currents ob-
tained from the model follows the same trend in each cycle. Figure 4.21 shows the
fitting of the PDC in PP2. The fitting of the polarization currents is quite accurate after
20 s, which means that the behaviour of G(t) has been reconstructed correctly. In the
first 20 s the fitting is not accurate: this matter will be discussed in section 4.9. In the
depolarization plot only the fit of the longest depolarization is shown, because all the
currents have the same trend in each cycle. Table 4.3 reports the free parameters related
to the physical model for PP1 and PP2. PP2 was the only sample with the appropriate
PDC measurements to fit the value of τa .

67
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

Table 4.3: Fitting parameters of PP1 and PP2 (ε0 is the dielectric permittivity of vacuum).

Parameter PP1 PP2

ε∞ [F/m] 2.6ε0 2.6ε0
σ1 [S/m] 7.0 · 10−12 2.4 · 10−14
σ2 [S/m] 2.0 · 10−15 4.6 · 10−15
n 1.3 0.3
τa [s] – 6.0 · 103
N [C/m3 ] 1.7 0.1
α [eV] 0.5 0.75

Polarization Depolarization

Figure 4.20: Reconstruction of the polarization and depolarization transients in PP1.

Polarization Depolarization

Figure 4.21: Reconstruction of the polarization and depolarization transients in PP2.

68
 4.8. Fitting of PDC Measurements

4.8.2 PP3 Transient Reconstruction

When materials are tested at different temperatures their properties will generally change.
As a consequence, some of the model parameters are expected to exhibit a certain de-
pendence on temperature.
Table 4.4 reports the parameters of PP3 at each temperature. G1 , G2 and n exhibit
a significant sensitivity to temperature. There may be a temperature dependence also
for τa , but the lack of suitable experiments for PP3 prevented the retrieval of this infor-
mation. The fitting of PDC curves is reported in Figure 4.22. The polarization fitting
is acceptable after approximately 20 s at all temperatures. There is a deviation between
the experiment and the fitting occurring between 5 · 103 s and 2 · 104 s at 30 ◦ C, and
a similar one occurs at 90 ◦ C, between 2 · 102 s and 4 · 104 s. The deviations between
fittings and experiments will be discussed in section 4.9.

Table 4.4: Fitting parameters of PP3 (ε0 is the dielectric permittivity of vacuum).

Parameter 30 ◦ C 60 ◦ C 90 ◦ C
ε∞ [F/m] 2.6ε0 2.6ε0 2.6ε0
−13 −13
σ1 [S/m] 2.6 · 10 1.6 · 10 1.1 · 10−12
σ2 [S/m] 1.3 · 10−16 1.2 · 10−14 9.2 · 10−14
n 1.1 0.4 0.3
3
N [C/m ] 0.1 0.1 0.1
α [eV] 0.75 0.75 0.75

The permittivity, ε∞ , is generally not dependent on temperature, at least in the op-

erating range of an insulating material. The parameters of the deep traps, N and α,
guarantee an acceptable fitting with the same value at all temperatures. All the fitted
depolarization curves in Figure 4.22 are very close to each other: temperature does not
have a significant impact in the model. The fitting seems acceptable for currents at
90 ◦ C and 60 ◦ C. The spike occurring at 2 · 103 s in the current at 60 ◦ C was likely due
to a disturbance during the measurement. The fitting at 30 ◦ C is acceptable after 103 s.

Polarization Depolarization
Current [A]

Current [A]

Time [s] Time [s]

Figure 4.22: Reconstruction of the polarization and depolarization transients in PP3.

69
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

Polarization Depolarization

∞ ∞

Figure 4.23: Simulated polarization and depolarization currents components in PP3.

The simulated currents can help to understand which component gives the major
contribution to the measured current, im (t), in each part of the transient. As a refer-
ence, Figure 4.23, shows the components of the PDC measured in PP3 at 60 ◦ C. The
first 20 s of the depolarization transient are dominated by the current associated with
fast polarization processes, if p (t) = C∞ v̇s (t). After 20 s the escape (detrapping) cur-
rent rules the discharging process, Ḃ = −ie (t). In the polarization transient the first
20 s of the simulated current are still dominated by if p (t), even if the model is not sup-
ported by the experimental data. For longer times the drift current, idr (t) = G(t)vs (t),
prevails and the agreement with the experiment is good. The fast polarization current,
if p (t), is the same in both polarization and depolarization of Figure 4.23. This is not
evident from the pictures because of the different axis scales. This current is due to the
capacitive branch of the circuit in Figure 4.14, which behaves symmetrically in both po-
larization and depolarization because the time constant of the generator was assumed
not to change between polarization and depolarization.

4.8.3 XLPE Transient Reconstruction

Table 4.5 summarizes the fitting parameters of the XLPE sample. This time the R-
C branch with its parameters τsp and Csp was necessary because the fast polarization
current alone was not sufficient to accurately reproduce the first 20 s of the depolar-
ization transients, as illustrated by Figure 4.24. Figure 4.25 shows the fitting of the
PDC in XLPE. The fitting of the polarization is not particularly accurate within the first
20 s. For times longer than 20 s the measured current at 20 ◦ C exhibits an oscillation
compared to the corresponding fitting: the experimental current lies below the fitted
one between 20 s and 5 · 103 s, and above between 5 · 103 s and 104 s. At 60 ◦ C and at
90 ◦ C the experimental current lies below the fitted one between 40 s and 6 · 103 s. The
fitting of the depolarization currents is very good at all temperatures in the first 20 s
of the transient. Beyond the first 20 s different deviations from the experiments occur.
At 20 ◦ C the fitting is good between 20 s and 200 s, then the experimental current de-
viates slightly above the fitting one. At 60 ◦ C the experimental current lies above the
fitting one between 20 s and 100 s, then the two curves reconcile with each other and
are in good agreement. The fitted current at 90 ◦ C is overlapped to the one at 60 ◦ C. In

70
 4.8. Fitting of PDC Measurements

this case the deviation between model and experiment is enhanced, being the measured
current above the fitted one between 100 s and 103 s.
Table 4.5: Fitting parameters of XLPE (ε0 is the dielectric permittivity of vacuum).

Parameter 20 ◦ C 60 ◦ C 90 ◦ C
ε∞ [F/m] 2.3ε0 2.3ε0 2.3ε0
τsp [s] 3.5 3.5 3.5
−12 −12
Csp [F] 8.0 · 10 8.0 · 10 8.0 · 10−12
σ1 [S/m] 9.4 · 10−14 1.1 · 10−13 1.5 · 10−14
σ2 [S/m] 2.8 · 10−17 1.5 · 10−15 1.5 · 10−14
n 0.9 0.6 0.5
N [C/m3 ] 0.3 0.3 0.3
α [eV] 0.5 0.5 0.5

Figure 4.24: Fitting improvement after introducing an R-C branch to take into account the slow polar-
ization processes.

Polarization Depolarization
Current [A]

Current [A]

Time [s] Time [s]

Figure 4.25: Reconstruction of the polarization and depolarization transients in PP3.

71
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

Overall, the presented results show that the model, within some limitations, pro-
vides an acceptable fitting of the experimental results. Deviations between model and
experiments are discussed in the next section.

4.9 Concluding Comments

The presence of space charge in the samples may cause the discrepancy observed be-
tween the fitting currents and the measured ones. To verify this idea, the space charge
density is estimated using a modified version of the formula derived in (4.50):
Z t
1
ρs (t) = im (t) − imod (t) dt. (4.86)
Ad t0
Time t0 is the instant at which the first current measurement occurs, and it is typically
greater than 0 s. To support the idea that space charge can be involved in the mis-
fitting, pulsed electroacoustic (PEA) measurements, [46], were performed on similar
specimens to those used for Figure 4.22 and Figure 4.25, namely PP3 and XLPE. The
specimens were subject to an electric field of 30 kV/mm for 104 s at two temperatures
levels, 30 ◦ C and 60 ◦ C for PP3, 20 ◦ C and 60 ◦ C for XLPE. Then the voltage was re-
moved. The space charge considered here is the net one, obtained as the net balance
between positive and negative charges. Charge ρs was then compared to the measured
one, ρm . As an example, Fig. 19 shows the match between the space charge measured
from XLPE at 20 ◦ C and its estimate from (4.86).
Table 4.6 compares ρs and ρm for PP3 and XLPE at the end of the polarization
transient, i.e., 104 s. The values do not perfectly match, but they are in the same order
of magnitude. This suggests a correlation between space charge and the discrepancy in
the fitting of the polarization currents.
Table 4.6: Comparison between measured (ρm ) and estimated (ρs ) space charge at 104 s. The value of
ρm was obtained by averaging the latest 100 experimental measurements.

Space charge at 104 s [C/m3 ]

Sample
Measured, ρm Esimated, ρs
◦
PP3 30 C 0.44 0.62
◦
PP3 60 C 0.11 0.61
◦
XLPE 20 C 0.80 0.77
XLPE 60 ◦ C 0.84 2.3

Space charge may also be involved in the initial stage of the depolarization transient.
The charge here should be stored in the shallow traps, thus it is ready to leave the
dielectric immediately after the beginning of the depolarization. Figure 4.27 displays
ρm obtained from the PEA measurements during the depolarization of PP3 at 60 ◦ C
[46]. The first 20 s of the transient are affected by a non-negligible amount of fast-
released (shallow-trapped) charge. This influences the behaviour of the current in the
initial part of the depolarization transient and hence the fitting.
Space charge may not be the only factor causing a discrepancy between model and
experiments. As shown in the polarization transients of Figure 4.20, Figure 4.21, Fig-

72
 4.9. Concluding Comments

Figure 4.26: Comparison between estimated (ρs ) and measured (ρm ) space charges in XLPE at 20 ◦ C
during polarization.

Figure 4.27: Space charge transient measured in PP3 at 60 ◦ C during depolarization.

ure 4.22 and Figure 4.25, within the first seconds of the polarization transient the fitting
is not good. The generator intrinsic transient, parasitic capacitances in the measure-
ment system and a delay between the moment in which the voltage is applied to the
sample and the one in which the measurement system starts the acquisition may have
a non-negligible effect. All these factors render the fitting of the initial polarization
and depolarization transients a challenge. As an example, let us consider PP3 at 60 ◦ C,
that was shown in Figure 4.23. The initial depolarization transient was properly fitted,
but not the polarization transient. If the model for the generator during polarization is
slightly modified by introducing a time-shift, ts ,
 
t−ts
vs (t) = V 1 − e τg
, (4.87)

with ts = −12 s, the fitting of the polarization curve becomes the one of Figure 4.28.
The fitting of the depolarization does not change with respect to Figure 4.23. Compared
to the polarization fitting in Figure 4.23, the one in Figure 4.28 is much better, but it
carries the cost of a new parameter, ts , due to the previously mentioned disturbing
factors. It is important to observe that the introduction of this new parameter did not
require to modify those associated with the shallow and the deep traps, that were listed

73
Chapter 4. Trapping and Detrapping in Polymers: a Circuital Model

Figure 4.28: Fitting of the polarization current in PP3 at 60 ◦ C after introducing the time shift ts .

in Table 4.4. This means that even if many disturbances occur to the measurement in
the first tens of seconds, the long term transient is insensitive to them, thus the fitting of
shallow and deep traps behaviour is robust. Different experimental setups from the one
used for the presented results may attenuate some of the disturbing factors in the initial
stage of the transients, allowing a better fitting of the model parameters over the whole
testing time.

74
 CHAPTER 5
Partial Discharges Modelling: the ks Approach

This chapter covers the modelling of partial discharges in insulation sys-

tems, a parallel topic with respect to the one treated in the previous part
of this work. The chapter takes two typical problems in partial discharge
modelling, namely the calculation of the partial discharge inception voltage
(PDIV) and the study of the inception of discharges in an internal cavity of
a polymer, and proposes two alternative modelling approaches with respect
to those traditionally employed to tackle these problems. The chapter be-
gins with a brief overview on the physics of discharges in air, in which the
Meek criterion is recalled, a fundamental tool in partial discharges mod-
elling. Then, after reviewing the traditional Niemeyer approach for PDIV
calculation, an alternative and more general method that has been developed
by the author of this work, called the ks approach, is presented. Afterwards,
a case study concerning the inception of surface discharges in an insulation
system demonstrates the effectiveness of the proposed ks approach. Finally,
the problem of discharge inception in the internal cavity of a polymeric mate-
rial is treated first with a simple RC model, which is the traditional approach
used in literature, and then with the polymer model developed in the previous
part of this work. The different predictions of the two models are presented
and discussed.

5.1 The Origin of a Discharge

Figure 5.1 shows a conceptual configuration in which two surfaces, Σa and Σb , are
charged with charges of opposite sign. The surfaces are separated by a volume of air in
between and above them, while on the bottom they are both touching a third surface, Σc ,
without charge. An electric discharge occurring between Σa and Σb can be classified

75
Chapter 5. Partial Discharges Modelling: the ks Approach

Γ1

_ _ _ _ _ _
+
+
Σa Σb

+
+
+
Γ2

+
Σc
Figure 5.1: Conceptual representation of a possible configuration that gives rise to electric discharges.
Surfaces Σa is positively charged, surface Σb is negatively charged, while surface Σc (dashed) is
neutral. Γ1 and Γ2 represent two possible paths along which the discharge can take place.

according to its path: if the path is Γ1 then it is a discharge in air, if the path is Γ2 it
is a surface discharge. The discharge can be further classified into a full discharge, or
flashover, if it bridges surfaces Σa and Σb , otherwise it is a partial discharge.
The discharge process begins with a gas of very low electrical conductivity exposed
to a high electric field. This chapter will treat only the case of air, in which the conduc-
tivity is mostly due to the presence of free electrons in the volume. The electric field
accelerates the free electrons already present in air, that will experience different types
of collisions with the neutral gas molecules (see Figure 5.2).
Elastic collisions. A portion of the electron’s kinetic energy is transferred to the molecule’s
kinetic energy. The total kinetic energy is conserved.
Excitations. A portion of the electron’s kinetic energy is converted into rotational ki-

e-
-
e -ΔEk

+ΔEk
Elastic collisions

Excitation of the e-
Small number gas molecule e-
Low σ -ΔEk
of free e-

Accelleration Collisions with

High external field +ΔEk = ΔEr or ΔEp
of free e- neutral gas molecues
e-
-
e
Ionization

Ion (+)
e- +
Attatchment
e-

_
Ion (-)

Figure 5.2: A schematic representation of the electrons’ collisions with molecules in air: σ is the electric
conductivity, e− is a free electron, ∆Ek is a variation of kinetic energy, ∆Er is a variation of
rotational kinetic energy and ∆Ep is a variation of potential energy. The dashed arrows on the far-
right portion of the scheme represent the trajectories of the colliding free electron. The solid arrows
represent the effect on the colliding molecule.

76
 5.1. The Origin of a Discharge

netic energy, or vibrational potential energy, or it is used to excite an electron to a

higher energy level of potential energy.
Ionization. The electron’s kinetic energy transferred to the molecule is used to free
an electron in the valence band. After this process, the molecule has become a
positive ion, and a new free electron is available in the volume.
Attatchment. The electron loses all its kinetic energy and falls into the valence band
of the molecule. After this process, the molecule has become a negative ion, and
one less free electron is available in air.
The possibility of triggering a discharge is determined by the number of free elec-
trons available in air, hence the ionization and attachment collisions are crucial, as they
immediately impact on the number of free electrons. This idea can be quantified by
defining an ionization coefficient, α, and an attachment coefficient, η, which represent
the number of electrons freed after or captured after a collision with a gas molecule.
Their difference is called effective ionization coefficient,

ᾱ = α − η, (5.1)

which determines the net growth or decay of the number of free electrons in air, n, by
means of the following differential equation (ds is a distance along a field line) [47]:

dn = nᾱ ds. (5.2)

The effective ionization coefficient is in general field- and pressure-dependent:

(
ᾱ(E, p) > 0 if E > Ecr
, (5.3)
ᾱ(E, p) = 0 if E ≤ Ecr

where p is the gas pressure and Ecr is usually referred to as the critical field, which
represents the minimum field required to have a net growth of free electrons. Equation
(5.2) can be integrated by separation of variables over a field line Γ:
Z n Z
dn
= ᾱ(E(s), p) ds, (5.4)
n0 n Γ

where n0 is the number of free electrons available at the beginning of line Γ and it was
assumed that pressure is constant, which a reasonable assumption. If ᾱ(E) > 0 then
the number of free electrons grows exponentially, [47, 48],
Z 
n = n0 exp ᾱ(E(s), p) ds . (5.5)
Γ

The exponential growth of free electrons is called avalanche. The avalanche is the
precursor of the discharge, which will take place only if the number of free electrons
overcomes a critical value, Ncr :
Z 
n0 exp ᾱ(E(s), p) ds ≥ Ncr . (5.6)
Γ

77
Chapter 5. Partial Discharges Modelling: the ks Approach

Taking the natural logarithm of both sides in (5.6) and considering that ln n0 is negli-
gible with respect to both ln Ncr and the integral term, the critical avalanche criterion,
or Meek criterion, is obtained [10, 47, 48]:
Z
ᾱ(E(s), p) ds ≥ Kcr , (5.7)
Γ

where Kcr = ln Ncr is called the Meek number.

The Meek criterion is the starting point of the models discussed in this chapter. The
general problem can be stated in the following manner.
Starting from an assigned geometric configuration of an electric system and
given a certain voltage waveform applied to the conductive parts, the electric
field evolution in space and time can be calculated. The model has to answer
the following question: which is the minimum voltage that is sufficient to
trigger a partial discharge in the insulation bulk or on its surface, according
to the Meek criterion?
The minimum voltage that triggers a partial discharge is, indeed, the PDIV.

5.2 Niemeyer’s Approach

L. Niemeyer proposed in [10] a powerful approach that enables the calculation of the
PDIV for a great variety of insulation systems, that can be applied to both discharges
in air and on the surfaces of the insulating and conductive components. The model is
based on the following assumption.
Discharge along the path of maximum field intensity Among all the possible paths
Γ in (5.7), Niemeyer considered only one line along which the electric field has
its maximum intensity. This assumption has two important implications.
1. The problem reduces from a three-dimensional (for PD in air) or a two-
dimensional (for surface PD) geometry to a one-dimensional geometry.
2. The problem is shifted from a stochastic context to a deterministic one. In-
deed, the origin of a PD strongly depends on the initial distribution of free
electrons as soon as voltage is applied, which is an intrinsically statistical
quantity. Excluding all the possible lines except the one of maximum field
intensity implicitly assumes that there is a sufficient number of free electrons
along that path to trigger a discharge before anywhere else.
The electric field along the line Γ of maximum intensity can be written as, [10],
E(s) = Ep e0 (s) on Γ, (5.8)
where Ep is the peak electric field on the line and e0 (s) is the normalized field profile.
In order to employ the Meek criterion an expression for the effective ionization
coefficient, ᾱ, is necessary. Niemeyer suggested the following empirical law, [10],
 
     β
 C E − E

p if E > Ecr
p p
ᾱ(E, p) = cr , (5.9)


 0 if E ≤ E cr

78
 5.2. Niemeyer’s Approach

where C and β are free parameters used to fit the experimental behaviour of ᾱ. Equation
(5.9) uses the ratio (E/p), which is called pressure-reduced field. In particular,
 
E Ecr
= (5.10)
p cr pcr
is the pressure-reduced critical field, which is the ratio between the critical field and
the reference pressure at which it is measured, pcr . Equation (5.10) indicates that the
critical field is pressure-dependent. Niemeyer et al indicated in [49] an expression that
relates the critical field to the gas density, ng , and the temperature, T :
!
E
Ecr = f (T ) ng , (5.11)
ng
cr0

where (E/ng )cr0 is the reduced critical field at standard temperature and pressure and
f (T ) is a function that takes into account the temperature-dependence of the critical
field. At room temperature function f (T ) is equal to 1 [10]. Thus, assuming that the
gas is ideal, which is a good approximation for air, the well-known law
p
ng = (5.12)
kB T
allows to write the critical field in terms of the pressure-reduced critical field
 
E
Ecr = p. (5.13)
p cr0
Equations (5.8) and (5.9) can be substituted into (5.7) to obtain a more explicit
version of the Meek criterion. Let l be the length of the line of maximum field intensity,
Γ, and let ξ = s/l be the reduced curvilinear abscissa. The electric field profile along
Γ is qualitatively represented in Figure 5.3, where ξ = 0 corresponds the starting point
of Γ on surface Σa in Figure 5.1 and ξ = 1 corresponds to the ending point of Γ on
surface Σb on Figure 5.1. Starting from ξ = 0 and moving to the right, the electric field
has a first peak and then it decreases to an almost constant value until ξ = 0.5, which is
midway along Γ. For ξ > 0.5 the field intensity begins to rise again, reaching a second
peak in proximity of surface Σb . The field profile E(ξ) is strongly influenced by the

E(ξ)

Ecr

0 ξcr,a 0.5 ξcr,b 1 ξ

Figure 5.3: A qualitative representation of the field profile on the normalized axis ξ = s/l. The field is in
general expected to rise in proximity of charged interfaces (positions ξ = 0 and ξ = 1 corresponding
to surfaces Σa and Σb in Figure 5.1, respectively). Far away from the charged interfaces the field is
almost uniform, and in ξ = 0.5 reaches it its minimum value.

79
Chapter 5. Partial Discharges Modelling: the ks Approach

shape of surfaces Σa and Σb , which in general might not have the same geometry, hence
the two peaks in Figure 5.3 are in general different. The effective ionization coefficient
is non-zero only when E(ξ) > Ecr , and the intervals in Figure 5.3 where this condition
holds are [0, ξcr,a ) and (ξcr,b , 1]. If the field profiles in the two intervals are different then
the Meek criterion must be applied to each of them separately, and two different PDIV
will be derived. It is possible to focus only on one of the two intervals. Therefore let
us consider [0, ξcr,a ), and let us put, for the sake of generality, ξcr,a = ξcr . The effective
ionization coefficient in this interval can be written as
 β  β
E Ep e0 (s)
ᾱ(E(s), p) = C − 1 p, (5.14)
p cr Ecr
and substituting this expression into the general Meek criterion, (5.7), yields, after the
change of variable ξ = s/l and a few simple algebraic manipulations,
Z ξcr  β
Ep e0 (ξ) Kcr
− 1 dξ ≥ . (5.15)
0 Ecr C (E/p)βcr (pl)
The inequality (5.15) is the pivot of Niemeyer’s approach. The normalized field profile
e0 (ξ) is assumed to be known, and the goal is to calculate the minimum value of Ep that
satisfies (5.15). Once such an Ep is known, the PDIV can be calculated as
Z l
PDIV = V0 + Ep e0 (s) ds, (5.16)
0

where V0 is an integration constant depending on the system configuration (the integral

computes the voltage drop along Γ, however there might be a voltage bias depending on
which of the conductive surfaces are energized and which are connected to the ground).
To calculate Ep Niemeyer replaces the inequality in (5.15) with an equal sign and
defines a generic function F as
Ep
= F (pl, gas-surface interface, insulation material, geometry). (5.17)
Ecr
Function F is always well-defined, because the integral in (5.15) is a monotonic func-
tion of the ratio Ep /Ecr and thus it can be inverted. Of course the expression of F can
rarely be calculated analytically, with the exception of few simple cases. A particularly

E(ξ)
Ep
Ecr

0 1=ξcr ξ
Figure 5.4: Uniform field distribution, obtained with e0 (ξ) = 1. In order to have a discharge the peak
field, Ep , must be greater than the critical field, Ecr . This implies that the critical distance, ξcr ,
extends to all the line length, thus ξcr = 1.

80
 5.3. The ks Approach

relevant one is the case of uniform field. In this situation ξcr = 1 and the normalized
field profile is e0 (ξ) = 1, as shown by Figure 5.4. The lines of maximum field inten-
sity in a configuration like that of Figure 5.1 are all those parallel to Γ2 . Niemeyer’s
inequality (5.15) becomes
 β
Ep e0 (ξ) Kcr
−1 ≥ , (5.18)
Ecr C (E/p)βcr (pl)
which yields
1
Ep (Kcr /C) β B
=F =1+ 1 = 1 + 1 , (5.19)
Ecr (E/p)cr (pl) β (pl) β
with
1
(Kcr /C) β
B= . (5.20)
(E/p)cr
In all those cases where analytic expression does not exist F can always be computed
numerically. However, its behaviour can be complicated, as it depends on many differ-
ent factors. It is possible to follow a simpler path, that will be illustrated in the next
section and has been published by the author of this work in [11].

5.3 The ks Approach

The main limitation of Niemeyer’s approach is that the F function is often difficult to
be computed, even numerically. The core of the problem lies in the inversion of the
integral
Z ξcr  β
Ep
e0 (ξ) − 1 dξ (5.21)
0 Ecr
to obtain an explicit value for the ratio Ep /Ecr . It is possible, however, to simplify
the approach by replacing F with another function, ks , that depends only on the field
profile. In general terms,
ks (Ep , geometry). (5.22)
The dependence on the geometry is enclosed by the normalized field profile, e0 (ξ). The
fact that ks depends on less parameters than F makes the approach more general. The
objective of the ks approach is to calculate the minimum value of Ep that triggers a
discharge, according to the Meek criterion. Let us write ks simply as ks (Ep ), to remark
the fact that Ep is the unknown of the problem. It is possible to define ks (Ep ) as
R ξcr h Ep e0 (ξ) iβ
0 Ecr
− 1 dξ
ks (Ep ) = h iβ . (5.23)
Ep
Ecr
−1

This definition places the integral (5.21), that appears on the left hand side of Niemeyer’s
inequality (5.15), at the numerator, while at the denominator it has the solution of the
same integral under the assumption of uniform field (e0 (ξ) = 1 and ξcr = 1). Function

81
Chapter 5. Partial Discharges Modelling: the ks Approach

ks (Ep ) yields is a positive number that measures how far the field distribution is from
the uniform case. More specifically:
(
ks (Ep ) = 1 with uniform field,
(5.24)
ks (Ep ) < 1 with non-uniform field.
Niemeyer’s inequality can be rewritten in terms of ks (Ep ) by multiplying and diving
 β
the left hand side of (5.15) by Ep /Ecr − 1 and then using (5.23):
h iβ
 β R ξcr Ep e0 (ξ) − 1 dξ
Ep 0 Ecr Kcr
−1 · iβ ≥ , (5.25)
Ecr C (E/p)βcr (pl)
h
Ep
Ecr
− 1
 β
Ep Kcr
− 1 · ks (Ep ) ≥ . (5.26)
Ecr C (E/p)βcr (pl)
Solving (5.26) for the ratio Ep /Ecr and using the definition of B, (5.20), yields, after
some manipulation,
Ep B
≥1+  1 . (5.27)
Ecr pl k (E ) β
s p
This inequality has a non-linear dependence on Ep , thus it can be solved only numeri-
cally. Up to this point, therefore, there is no evident advantage in using ks (Ep ) in place
of Niemeyer’s F function. The non-linearity in (5.27) can be overcome by removing
the dependence of ks on Ep : this requires, of course, a reasonable approximation. Con-
sider the electric field profile in Figure 5.5, which is shown in the range ξ ∈ [0, 0.5]. It
is possible to foster the following argument.
The portion of field profile above the critical field, Ecr , that contributes for
the major part to the integral in the definition of ks (Ep ), (5.23), is centred
around the peak of the profile itself. In order to simplify the calculation of
ks , then, it is possible to replace the field profile, E(ξ), with a rectangular
profile, Er (ξ), with the same peak value of E(ξ), and contained within a
suitable interval [ξ1 , ξ2 ].
The rectangular field profile can be written in terms of the Heaviside step function, H,
 
Er (ξ) = Ep H(ξ − ξ1 ) − H(ξ − ξ2 ) . (5.28)
If the rectangular field profile is used in place of the original one, then ks becomes
independent from Ep :
R ξ2 h Ep iβ
ξ1 Ecr
− 1 dξ s2 − s1
ks ≃ iβ = ξ2 − ξ1 = , (5.29)
l
h
Ep
Ecr
−1

where in the last expression the definition ξ = s/l was used. With this result the
inequality (5.27) is rewritten as
Ep B
≥1+ 1 . (5.30)
Ecr (plks ) β
82
 5.4. Surface Partial Discharge

E(ξ)
Er(ξ)
Ecr

0 ξ1 ξ2 0.5 ξ
Figure 5.5: Representation of the field profile, E(ξ), and of the rectangular field profile, Er (ξ), which
replaces the former in the approximate calculation of ks .

The minimum value of Ep , Ep,min , which is called inception field, Einc , is then calcu-
lated by replacing the ≥ sign with an equality in (5.30):
" #
B
Ep,min = Einc = Ecr 1 + 1 (5.31)
(plks ) β

The next two sections illustrate two case studies of PDIV calculation. The case
studies use the properties of the XLPE material investigated in chapter 4 during the
validation of the polymer circuital model. Two different types of discharges will be
investigated: a partial discharge on the surface of the material, which is driven by a
non-uniform electric field, and an internal partial discharge, simulated with a cavity
embedded in the material and driven by a uniform electric field.

5.4 Surface Partial Discharge

Figure 5.6 depicts the 2D geometric model used to simulate the conditions for the
partial discharge inception. The HV electrode is energized with a positive voltage,
while the GND electrode is fixed at zero potential. The simulation calculates the electric
field, E(s), on the sample surface. In particular, the field is calculated along the shortest
path, on the surface, that connects the triple points THV and TGN D . The electrodes are
never perfectly sharp at the triple point: they can be modelled by a curved boundary
with a certain curvature radius, ρ. For the current example a radius of 100 µm has been
chosen [11]. The geometric model is solved for the steady-state continuity equation

l
HV GND
E(s)
1.4 mm THV TGND Sample
ρ

ρ = 100 μm
Figure 5.6: Geometric model for the surface discharge. Length l is the distance between the electrodes.
THV and TGN D are the triple points between electrode, sample and air.

83
Chapter 5. Partial Discharges Modelling: the ks Approach

Figure 5.7: Temperature-dependence of the conductivity. The dashed line is the fitting, the dots are the
measured values. The slope, α, and the field at which the measurements were taken are indicated.

Figure 5.8: Field-dependence of the conductivity. The dashed line is the fitting, thedots are the measured
values. The slope, β, and the temperature at which the measurements were taken are indicated.

∇ · σ(T, E)E = 0, (5.32)
whose solution yields the resistive field profile determined by the temperature- and
field-dependent electric conductivity. Letting E = −∇ϕ, ϕ being the scalar potential,
the boundary conditions are, for a certain applied voltage V ,
(
ϕ = V on the HV electrode,
(5.33)
ϕ = 0 on the GND electrode.

The solution of equation (5.32) requires knowledge of σ(T, E), which for this ex-
ample is modelled as [23, 25]
 
σ(T, E) = σ0 exp α(T − T0 ) + β(|E| − E0 ) . (5.34)
For the XLPE material under investigation steady-state conductivity measurements
were available at different temperatures and fields, from which a suitable fitting of
σ0 , α and β was obtained. The results are summarized by Figure 5.7, Figure 5.8, and
Table 5.1.

84
 5.4. Surface Partial Discharge

Table 5.1: Parameters of the σ(T, E) model, (5.34).

Parameter Value
σ0 10−15 S/m
α 0.1 ◦ C−1
T0 60 ◦ C
β 0.1 mm/kV
E0 30 kV/mm

5.4.1 Simulation Results and PDIV Calculation

The solution of equation (5.32) with V = 1 V in the boundary conditions (5.33) at three
different distances, 5 mm, 25 mm and 50 mm, and three different temperatures, 30 ◦ C,
60 ◦ C and 90 ◦ C, yields the profiles shown in Figure 5.9, Figure 5.10 and Figure 5.11.
At 5 mm, Figure 5.9, the profile is uniform at 30 ◦ C−1 , and its value is 200 V/m. As
the temperature rises the field is no longer uniform, manifesting a rising peak close to

Figure 5.9: Electric field profiles at 30 ◦ C, 60 ◦ C and 90 ◦ C. Electrodes at 5 mm, applied voltage
V = 1 V.

Figure 5.10: Electric field profiles at 30 ◦ C, 60 ◦ C and 90 ◦ C. Electrodes at 25 mm, applied voltage
V = 1 V.

85
Chapter 5. Partial Discharges Modelling: the ks Approach

Figure 5.11: Electric field profiles at 30 ◦ C, 60 ◦ C and 90 ◦ C. Electrodes at 50 mm, applied voltage
V = 1 V.

the triple points (located at 0 mm and 5.2 mm).

At 25 mm, Figure 5.10, the profile is again uniform at 30 ◦ C, and its value is 40 V/m,
which is lower than the 200 V/m observed in Figure 5.9. When the temperature in-
creases sharper spikes than the case of 5 mm arise in proximity of the triple points
(now located at 0 mm and 25.2 mm).
At 50 mm, Figure 5.11, the profile is not completely uniform already at 30 ◦ C, hav-
ing a plateau midway between the triple points (at 0 mm and 50.2 mm). The field level
is close to 20 V/m, lower than the other cases. The spikes still rise with temperature.
The field-dependence in the conductivity makes equation (5.32) a non-linear prob-
lem. So, in principle, the field profiles shown in Figure 5.9, Figure 5.10 and Figure 5.11
depend on the voltage V applied through the boundary conditions (5.33). The question
is whether the sensitivity of the field profile with respect to the applied voltage is strong
or not. To verify this aspect Figure 5.12, Figure 5.13 and Figure 5.14 show the simu-
lation of the field profile at the three distances 5 mm, 25 mm and 50 mm, temperature
90 ◦ C and two applied voltages: 1 V and 100 kV. In all the three cases it is evident that
the profiles at different applied voltages are the same, which means that the field profile
is not significantly sensitive to the non-linearity of the problem, that can be, in practice,
treated as a linear one. This means that the field profile scales with the applied voltage
according to the proportionality law
E1 (s) V1
= , (5.35)
E2 (s) V2
being E1 (s) and E2 (s) the field profiles obtained at voltages V1 and V2 respectively.
Another important aspect is the electric conductivity of air, σa , to use in the sim-
ulation. In [50] a range between 10−14 S/m and 10−10 S/m is given, depending on
altitude. The results presented so far have been obtained with σa = 10−14 S/m, but it
should be kept in mind that the field profile is quite sensitive to this parameter. Fig-
ure 5.15 shows four profiles with the electrodes 5 mm apart. The first profile, a black,
thick line, is at 30 ◦ C and σa = 10−14 S/m, which was already shown in Figure 5.9 and
is reported here for reference. The remaining profiles are at 30 ◦ C, 60 ◦ C and 90 ◦ C,
with σa = 10−12 S/m. Raising the air conductivity by two orders of magnitude in this

86
 5.4. Surface Partial Discharge

Figure 5.12: Electric field profiles at 5 mm and 90 ◦ C with two applied voltages: 1 V and 100 kV.

Figure 5.13: Electric field profiles at 25 mm and 90 ◦ C with two applied voltages: 1 V and 100 kV.

Figure 5.14: Electric field profiles at 50 mm and 90 ◦ C with two applied voltages: 1 V and 100 kV.

87
Chapter 5. Partial Discharges Modelling: the ks Approach

Figure 5.15: Electric field profiles at 5 mm and 30 ◦ C, 60 ◦ C and 90 ◦ C and with the air conductivity,
σa , being 10−12 S/m. The case at 30 ◦ C and σa = 10−14 S/m has been reported as well, for
comparison with the results obtained in Figure 5.9.

simulation removes the rising spikes at increasing temperature that were observed in
Figure 5.9. A proper estimate of σa , hence, is necessary to obtain realistic field profiles.

The PDIV calculation relies on the theory presented in sections 5.2 and 5.3. The
algorithm is simple and will now be illustrated. First of all, the inception field is cal-
culated with equation (5.31), which is reported here for convenience (note that Ecr has
been substituted with its definition (5.13), (E/p)cr p) :
  " #
E B
Einc = p 1+ . (5.36)
p cr (plks )1/β

In the case of surface discharges the parameters are [11]: (E/p)cr = 8 V Pa−1 m−1 ,
B = 4.3 Pa0.5 m0.5 , p = 101325 Pa and β = 2. The value of ks (Einc ) is obtained
through the definition (5.23). Once Einc is known, the PDIV can be obtained through
equation (5.35) as
Einc
PDIV = Vsim , (5.37)
Ep,sim
where Vsim is the voltage applied in the simulation and Ep,sim is the peak value of the
simulated field profile.
The value of ks in equation (5.36) can be calculated using (5.29), ks = (s2 − s1 )/l.
The values of s1 and s2 are derived from the simulated field profile as shown by Fig-
ure 5.16. In practice, s1 and s2 are chosen in such a way that
E(s1 ) = E(s2 ) = f Ep,sim , (5.38)
where f is a factor between 0 and 1. The value of f is determined empirically and it
depends mostly on the geometry of the system and on the type of excitation, in this case
DC. For the configuration of Figure 5.1 a suitable value for f is 0.9.
Table 5.2 summarizes the PDIV calculated using both Niemeyer’s approach and the
ks approach. In the former case the values of the F function and of the inception field
are also reported. In particular, the inception field was calculated as Einc = F · Ecr ,

88
 5.4. Surface Partial Discharge

E(s)
Tangential field
Ep,sim Field approximation

f·Ep,sim

s1 s2 s
Figure 5.16: Illustration of the way factor f is employed to estimate s1 and s2 from the field profile.

where Ecr = (E/p)cr p ≃ 0.811 kV/mm. In the second case the values of ks and Einc
are reported, and the latter was calculated with (5.36). The value of Ep,sim was the
same in both approaches, as it comes from the calculation of the field profile. The field
profiles from which the PDIV is retrieved are those shown in Figure 5.9, Figure 5.10
and Figure 5.11. As it can be observed, both approaches yield similar values for both
Einc and the PDIV. This fact supports the ks approach as a viable tool for the estimation
of the PDIV.
Table 5.2: Parameters calculated from the profiles of Figure 5.9, Figure 5.10 and Figure 5.11. Tem-
peratures are in ◦ C, distances in mm, electric fields Ep,sim and Einc in kV/mm, and the PDIV in
kV. The critical field, Ecr , that links Einc to function F in Niemeyer’s approach, is 0.811 kV/mm.
The value of Ep,sim was obtained with Vsim = 1 V. A factor f = 0.9 was used in the approximate
approach.

Niemeyer’s approach ks approach

Temperature Distance Ep,sim
F Einc PDIV ks Einc PDIV
30 5 0.200 1.28 1.03 5.17 0.470 1.03 5.16
60 5 0.210 1.35 1.09 5.20 0.476 1.03 4.90
90 5 0.405 2.10 1.71 4.21 0.0174 1.96 4.83
30 25 0.0406 1.14 0.923 22.7 0.494 0.908 22.4
60 25 0.0559 1.50 1.21 21.7 0.0160 1.35 24.2
90 25 0.140 2.14 1.73 12.5 0.00346 1.98 14.2
30 50 0.0206 1.12 0.905 43.9 0.497 0.880 42.7
60 50 0.0214 1.16 0.938 43.7 0.422 0.886 41.3
90 50 0.0380 1.60 1.29 35.2 0.00583 1.45 39.4

5.4.2 Discussion
Figure 5.17 shows the behaviour of the PDIV, calculated with Nimeyer’s approach and
with the ks approach, as a function of the distance between the electrodes. It can be
observed that the data points at 30 ◦ C and 60 ◦ C follow a linear trend, meaning that
the PDIV increases linearly with the distance between the electrodes. The PDIV at
90 ◦ C does not follow the same linear trend, but it is rather an exponential one, as
demonstrated by Figure 5.18, where the data points at 90 ◦ C line up onto a straight

89
Chapter 5. Partial Discharges Modelling: the ks Approach

Figure 5.17: PDIV calculated with Niemeyer’s approach (Ni.) and with the ks approach (ks ) at 30 ◦ C,
60 ◦ C and 90 ◦ C, and as a function of the distance between the electrodes. Data is taken from Table
5.2. The PDIV grows linearly (dashed black line) with distance at 30 ◦ C and 60 ◦ C. At 90 ◦ C the
trend is no longer linear.

Figure 5.18: PDIV at 90 ◦ C calculated with Niemeyer’s approach (Ni.) and with the ks approach (ks )
as a function of the distance between the electrodes. Data is taken from Table 5.2. The vertical axis
is in logarithmic scale, while the horizontal axis is in linear scale. The straight line trend suggests
an exponential growth of the PDIV with distance (dashed black line).

line on a log PDIV vs. distance plot. The different behaviour of the line at 90 ◦ C
may be ascribed to the XLPE conductivity at this temperature with respect to the air
conductivity. Indeed, consider the following facts.

• In these simulations the conductivity of air, σa , was set to 10−14 S/m.

• As it can be observed from Table 5.2, for all the simulated configurations the in-
ception electric field is never greater than 1.70 kV/mm. Thus, in first approxima-
tion, the conductivity of XLPE can be thought to be only temperature-dependent.

At 30 ◦ C and 60 ◦ C the ratio between the XLPE conductivity and the air conductivity is

σXLPE (30 ◦ C) σXLPE (60 ◦ C)

≃ 0.005 and ≃ 0.1, (5.39)
σa σa
90
 5.5. Internal Partial Discharge

Figure 5.19: PDIV calculated with Niemeyer’s approach (Ni.) and with the ks approach (ks ) at 5 mm,
25 mm and 50 mm, and as a function of temperature. Data is taken from Table 5.2. The PDIV de-
creases with temperature, and this is linked to the rising field peak observed in Figure 5.9, Figure 5.10
and Figure 5.11.

while at 90 ◦ C is
σXLPE (90 ◦ C)
≃ 2. (5.40)
σa
Therefore, when σXLPE ≪ σa the PDIV grows linearly with the distance between
the electrodes, while when σXLPE ∼ σa the PDIV grows exponentially with the dis-
tance. This is a rather unintuitive behaviour, nevertheless the ks approach has been
able to predict the PDIV correctly: the values are indeed very close to those calcu-
lated with Niemeyer’s approach (Table 5.2, Figure 5.17 and Figure 5.18). This means,
in particular, that the idea that the discharge is driven by the electric field around the
peak is robust, and it can be employed as guidance in the design of surface insulation
systems. Indeed, triple points between conductive parts, insulating parts and air arise
in many insulation systems, causing local field magnifications with profiles similar to
those shown in Figure 5.9, Figure 5.10 and Figure 5.11 that raise the risk of triggering
partial discharges. This last fact is evidenced by the plot of Figure 5.19. The PDIV at
the three different distances is plotted against temperature. The decreasing trend, com-
bined with the observation of a rising field peak with increasing temperature, (again,
see Figure 5.9, Figure 5.10 and Figure 5.11) is a clear manifestation of the fact that a
local field magnification lowers the PDIV, raising the risk of discharge.

5.5 Internal Partial Discharge

Figure 5.20 depicts, on the left, a slab of XLPE material of thickness d, with an in-
ternal cylindrical cavity of height h. This configuration can be conceptualized with the
scheme shown on the right of Figure 5.20, where the cavity is connected between nodes
b and c, the series bulk representing the cylindrical column of height d − h aligned with
the cavity is connected between nodes a and b and the remaining parallel bulk is con-
nected between nodes a and c. For an applied voltage source vs (t), the series bulk and
the cavity see, respectively, voltages vb (t) and vc (t) such that vb (t) + vc (t) = vs (t),
while the parallel bulk sees the total source voltage.

91
Chapter 5. Partial Discharges Modelling: the ks Approach

Series d-h
vb(t)
bulk
d Parallel d
vs(t) vs(t)
b bulk
h
vc(t) Cavity h

Figure 5.20: (Left) A slab of XLPE material of thickness d, with an internal cylindrical cavity of height
h. (Right) A conceptual scheme that represent the XLPE slab. The blocks geometrical height and
voltage drops ar indicated. This scheme is used as starting point for developing circuital models.

vb(t) Cb Rb
+
vs(t) b vs(t) Cp Rp

vc(t) Cc Rc

Figure 5.21: Equivalent circuit for the study of an internal cavity. Each block is modelled with a parallel
RC circuit.

The conceptual scheme in Figure 5.20 is the starting point for developing circuital
models that allow to calculate the voltage drop across the cavity, vc (t), derive the PDIV
and evaluate the risk of triggering internal partial discharges. The simplest equivalent
circuit, which is widely used in literature [9, 51], models every block in the conceptual
scheme with a parallel RC branch, as shown in Figure 5.21. Improved versions of
this model can include, for example, series RC branches to describe slow polarization
processes in the dielectric, as done in the Occhini model. In this specific case the
material is XLPE, hence the model developed in chapter 4 for polymers will be used to
describe the series bulk and parallel bulk blocks, while the cavity, which is supposed to
contain air, will still be a parallel RC block. The final circuit depicted in Figure 5.22,
then, is obtained.
The study of an internal discharge in the XLPE slab is conducted as follows. First,
the material is described with the traditional equivalent circuit of Figure 5.21. Then
the PDIV is calculated at three different temperatures: 20 ◦ C, 60 ◦ C and 90 ◦ C. The
electric conductivity is assumed to be only temperature-dependent. The same is done
using the equivalent circuit of the polymer, Figure 5.22. It is important to observe that
the two equivalent circuits are derived under the assumption of uniform electric field,
both in the bulk and in the cavity. Therefore, the inception field and the PDIV are
calculated under this assumption, which implies that Niemeyer’s approach and the ks
approach (with ks = 1) yield exactly the same result. The focus of the study, then, is to
investigate the different predictions of the two circuital models, rather than the PDIV
calculation procedure.

92
 5.5. Internal Partial Discharge

ic(t) ie(t)
Rsp
vb(t) G(t) C∞ ie,p(t)
ic,p(t)
+ Csp Rsp,p
vs(t) Sdt(B) vs(t) Gp(t) C∞,p
Csp,p
b
Sdt,p(B)
vc(t) Cc Rc

Figure 5.22: Equivalent circuit for the study of an internal cavity, where the series bulk and parallel
bulk blocks are described with the model developed in chapter 4 for the polymer.

5.5.1 PDIV Calculation with the Traditional RC Model

The PDIV calculation requires knowledge of voltages vb (t) and vc (t). Therefore, the
parallel block is irrelevant according to the circuit in Figure 5.21. The study then will
rely on the equivalent circuit of Figure 5.23, where only the relevant blocks are drawn.
The voltage source is modelled with Heaviside’s step function,

vs (t) = V H(t). (5.41)

The geometric parameters of problem are summarized by Table 5.3 (they were adapted
from [12]).

Table 5.3: Geometric parameters of the simulated system.

Description Variable Value

Cavity height h 50 µm
Cylinder base radius r 1.5 mm
Cylinder base area A = πr2 7.07 mm2
XLPE slab height d 0.5 mm

The circuital components of the bulk, Cb and Rb , and of the cavity, Cc and Rc , are then
obtained from the permittivity and conductivity of the bulk, εb and σb , and of the cavity,

vb(t) Cb Rb
+
vs(t) b

vc(t) Cc Rc

Figure 5.23: Equivalent circuit based on the RC model for the PDIV calculation.

93
Chapter 5. Partial Discharges Modelling: the ks Approach

εc and σc , as:
A 1 d−h A 1 h
C b = εb , Rb = , Cc = εc , Rc = . (5.42)
d−h σb A h σc A
The values of the microscopic and circuital parameters in (5.42) are listed in Table
5.4. The bulk conductivity is the state-state conductivity obtained from the fitting of
the XLPE measurements, more specifically it is the value of σ2 from Table 4.5. The
cavity conductivity, σc , has been taken equal to 10−14 S/m, which is the same of the air
conductivity employed for the surface discharge simulations.
Table 5.4: Microscopic and circuital parameters of the RC model in Figure 5.23.

Description Microscopic Value Circuital Value

Bulk permittivity εb [F/m] 2.3ε0 Cb [F] 3.2 · 10−13
Bulk conductivity at 20 ◦ C σb [S/m] 2.8 · 10−17 Rb [Ω] 2.3 · 1018
Bulk conductivity at 60 ◦ C σb [S/m] 1.5 · 10−15 Rb [Ω] 4.2 · 1016
Bulk conductivity at 90 ◦ C σb [S/m] 1.5 · 10−14 Rb [Ω] 4.2 · 1015
Cavity permittivity εc [F/m] ε0 Cc [F] 1.3 · 10−12
Cavity conductivity σc [S/m] 1.0 · 10−14 Rc [Ω] 7.1 · 1014

The circuit in Figure 5.23 is described by the following system of equations:

(
vs = vb + vc
. (5.43)
Cb v̇b + Rvbb = Cc v̇c + Rvcc

Substituting vc = vs − vb into the second equation yields an ordinary differential equa-

tion in vb :  
1 1 vs
(Cb + Cc ) v̇b + + vb = Cc v̇s + . (5.44)
Rb Rc Rc
Dividing the above equation by (Cb + Cc ) yields
vb Cc vs
v̇b + = v̇s + , (5.45)
τ Cb + Cc Rc (Cb + Cc )
where
Cb + Cc
τ= 1 . (5.46)
Rb
+ R1c
The solution of the homogeneous equation is Ke−t/τ , while the steady-state solution,
recalling the applied voltage waveform (5.41), is
τV Rb
vb (∞) = = V, (5.47)
Rc (Cb + Cc ) Rb + Rc
which underlines that the steady-state voltage is determined by the resistances, and thus
by the conductivities. The general solution of the problem, then, is
Rb
vb (t) = Ke−t/τ + V. (5.48)
Rb + Rc
94
 5.5. Internal Partial Discharge

To find the value of K an initial condition is necessary. To this aim, the general solution
is substituted into (5.45), keeping in mind that the derivatives have to be interpreted in
a distributional sense:
K
v̇b = − e−t/τ + vb (0+ ) − vb (0− ) δ(t)


(5.49)
τ
and
v̇s = vs (0+ ) − vs (0− ) δ(t),


(5.50)
where δ(t) is the Dirac delta function. Considering that vs (0+ ) = V , while vs (0− ) =
vb (0− ) = 0 V (there is no applied voltage for t < 0), equation (5.45) in t = 0+ , after
some simplifications, reduces to
Cc
vb (0+ )δ(t) = V δ(t), (5.51)
Cb + Cc
that yields the initial condition
Cc
vb (0+ ) = V. (5.52)
Cb + Cc
This last equation states that the initial voltage distribution is determined by the capac-
itances, and thus by the permittivities. In conclusion, substituting the initial conditions
into the general solution yields the value of K,
Cc Rb
K= V − V, (5.53)
Cb + Cc Rb + Rc
and the final solution, for t ≥ 0, is
"  #
Cc Rb R b
vb (t) = V − e−t/τ + . (5.54)
Cb + Cc Rb + Rc Rb + Rc

The cavity voltage can be obtained, for t ≥ 0, as vs (t) = V − vb (t). Observing that
Cc Rb Cb Rc
−1+1− =− + , (5.55)
Cb + Cc Rb + Rc Cb + Cc Rb + Rc
the final expression for vc (t) is
"  #
Cb Rc R c
vc (t) = V − e−t/τ + . (5.56)
Cb + Cc Rb + Rc Rb + Rc

The calculation of the PDIV requires knowledge of the inception electric field,
which can be derived, for example, from equation (5.36) with ks = 1 and l = h:
  " #
E B
Einc = p 1+ . (5.57)
p cr (ph)1/β
The use of this equation, however, demands special care in the case of an internal
discharge. In this study the material is evaluated at three different temperatures, namely

95
Chapter 5. Partial Discharges Modelling: the ks Approach

20 ◦ C, 60 ◦ C and 90 ◦ C. The cavity is embedded in the XLPE slab, therefore the gas
contained in it is at the same temperature of the bulk. In first approximation, it is
reasonable to assume that the gas density remains constant, therefore, according to the
ideal gas law, ng = p/kB T , if the temperature changes by a factor c, the pressure has
to change by the same factor:
p cp
ng = const. =⇒ = (ng kb ) = const. = . (5.58)
T cT
In the following, it will be assumed that the atmospheric pressure p = 101325 Pa
corresponds to the room temperature of 20 ◦ C. Therefore, the formula for the inception
field in case of the internal discharge is
  " #
E B
Einc = p 1+ , (5.59)
p cr (ph)1/β

where (E/p)cr = 25.2 V Pa−1 m−1 , B = 5.8 Pa0.5 m0.5 and β = 2 [10, 11], while
the value of p has to be selected according to the temperature. This argument was not
carried in the case of the surface discharge because the discharge took place in open air,
thus even though the sample was at different temperature levels, the air volume could
be approximated to a thermal reservoir: its temperature would not change and pressure
could be kept at the atmospheric level in all conditions.
Once the inception field has been obtained, the inception voltage across the cavity
can be calculated, under the assumption of uniform field, as

Vinc = Einc h. (5.60)

According to the definition given at the end of section 5.1, the PDIV is the minimum
applied voltage V such that

vc (t) ≥ Vsim ∀t ≥ 0. (5.61)

Figure 5.24: Evolution of the normalized cavity voltage, vc (t)/V , over time. The maximum voltage is
always obtained at t = 0, when the voltage is determined by the ratio of the capacitances, and it
is independent from temperature. At steady state instead the voltage is determined by the ration of
resistances, thus it is linked to the temperature-dependence of the conductivity.

96
 5.5. Internal Partial Discharge

Figure 5.24 shows the behaviour over time of the normalized cavity voltage, vc (t)/V . It
is evident that the peak voltage occurs at t = 0, therefore the PDIV can be determined
by solving the following equation for V :
 
+ Cb Cc
vc (0 ) = V = Vinc ⇔ V = Vinc 1 + . (5.62)
Cb + Cc Cb
Table 5.5 summarizes the PDIV calculation results using (5.62). A thorough discussion
of these results will be presented after calculating the PDIV using the polymer model.
Table 5.5: PDIV calculation results using equation (5.62).

T [◦ C] p [Pa] Einc [kV/mm] Vinc [kV] PDIV [kV]

20 1.01 · 105 9.13 0.457 2.24
60 3.04 · 105 19.1 0.953 4.68
90 4.56 · 105 25.4 1.27 6.25

5.5.2 PDIV Calculation Using the Polymer Model

As already explained in the case of the RC model, the PDIV requires knowledge of
voltages vb and vc , thus the parallel block can be removed from the electric circuit,
yielding, in the case of the polymer model, the one in Figure 5.25. In this model there
are four state variables: voltages vb and vc , voltage vsp from the polarization branch and
charge B stored in the non-linear capacitor Sdt (B). The system of equations is:


 vs = vb + vc
  
e −B − B


 Ḃ = G(t)v Se N
b τ (B)
= ic − ie
(5.63)


 v sp + τ sp v̇ sp = vb
 v G(t) + C v̇ + C v̇ + Ḃ = C v̇ + cv

b ∞ b sp sp c c Rc

which is a differential-algebraic system of equations (DAE) of index 1, that is solved

numerically. The parameters that appear in (5.63) were all defined in chapter 4. The
numerical solution of the model does not allow to use the Heaviside step function for

ic(t) ie(t)
Rsp
vb(t) G(t) C∞
+ Csp vsp(t)
vs(t) Sdt(B)
b

vc(t) Cc Rc

Figure 5.25: Equivalent circuit based on the polymer model for the PDIV calculation.

97
Chapter 5. Partial Discharges Modelling: the ks Approach

the applied voltage source, vs , which is, therefore, approximated by a fast-rising expo-
nential,  
vs (t) = V 1 − e−t/τg , (5.64)

with τg = 30 ms. In the simulation, then, one should expect vc (t) to start from the
origin at t = 0, quickly rise to a peak value and then slowly decay towards the steady
state condition. Table 5.6 summarizes the parameters used to run the simulations.

Table 5.6: Model parameters for the DAE (5.63).

Parameter 20 ◦ C 60 ◦ C 90 ◦ C
C∞ [F] 2.3ε0 2.3ε0 2.3ε0
τsp [s] 3.5 3.5 3.5
Csp [F] 1.3 · 10−13 1.3 · 10−13 1.3 · 10−13
G1 [S] 1.5 · 10−15 1.7 · 10−15 2.4 · 10−16
G2 [S] 4.4 · 10−19 2.4 · 10−17 2.4 · 10−16
n 0.9 0.6 0.5
−10 −10
N [C]
e 9.5 · 10 9.5 · 10 9.5 · 10−10
α [eV] 0.5 0.5 0.5
Se [C−1 ] 8.8 · 109 8.8 · 109 8.8 · 109
Cc [F] 1.3 · 10−12 1.3 · 10−12 1.3 · 10−12
Rc [Ω] 7.1 · 1014 7.1 · 1014 7.1 · 1014

The PDIV calculation is different from the RC model because of the non-linearity
of the polymer model. Figure 5.26 shows the normalized cavity voltage, vs (t)/V , cal-
culated at 60 ◦ C with two different applied voltages: 100 V and 1000 kV. The two
curves are different in the first part of the transient, while they come close as they reach
the steady-state. Such a strong dependence on the applied voltage magnitude V is a

Figure 5.26: Normalized cavity voltage, vs (t)/V , at 60 ◦ C with two different applied voltages: 100 V
and 1000 kV. The fact that the two curves do not overlap in every point indicates that the non-
linearity of the model is not negligible.

98
 5.5. Internal Partial Discharge

Figure 5.27: Simulated cavity voltages, vc (t), with applied voltage equal to the calculated PDIV from
Table 5.7.

symptom of non-negligible non-linearity, which differs from what had been found in
the case of the surface discharge. Therefore, being Vinc = Einc h the inception voltage
across the cavity, the PDIV is obtained from the numerical solution of the following
equation in V :
max vc (V, t) = Vinc , (5.65)
t≥0

where the explicit dependence of vc on V has been written for clarity. Table 5.7 com-
pares the PDIV obtained from the RC model and the ones obtained from the polymer
model (the other quantities that appeared in Table 5.5 remain the same). The PDIV
obtained with the polymer model is always smaller than the one obtained with the RC
model. Figure 5.27 shows the cavity voltage, vc (t), at the three temperatures when the
corresponding PDIV obtained from the polymer model is applied.

Table 5.7: Comparison of the PDIV obtained with the RC model and the polymer model.

T [◦ C] PDIV, RC model [kV] PDIV, polymer model [kV]

20 2.24 1.29
60 4.68 3.33
90 6.25 4.66

5.5.3 Discussion
The discussion carried so far had one goal: showing that the polymer model can provide
significantly different estimates of the PDIV with respect to the traditional RC model.
From the plots of Figure 5.24 and Figure 5.27, it is clear that the calculated PDIV
was referred to the initial stage of the transient. In the RC model this also called
the PDIVAC , underlining the fact that the voltage is determined by the ratio of the
capacitances, and thus of the permittivities (resistive phenomena are negligible in AC)
[52]. The same nomenclature, however, cannot be adopted with the polymer model,
because at the initial stage of the transient a significant contribution is given by the
current of the slow polarization process (Csp v̇sp ), the conduction through shallow traps

99
Chapter 5. Partial Discharges Modelling: the ks Approach

(vb G(t)) and through the deep traps (Ḃ): this is the reason of the differences in the
PDIV observed in Table 5.7.
At DC steady state however, the situation is different. In the polymer model, the
slow polarization process has terminated its transient, thus Csp v̇sp = 0 A; the deep
trapping phenomena have reached the dynamic equilibrium, hence Ḃ = 0 A; and the
conduction through shallow traps has reached the steady state, meaning that conduc-
tance G(t) is at its steady-state value. Therefore, the polymer model reduces to the
series of the bulk and cavity resistances, and it is equal to the steady-state RC model.
In this situation it is possible to define, for both models, the PDIVDC as
 
Rb
PDIVDC = Vinc 1 + , (5.66)
Rc
which has been obtained by equating the steady-state cavity voltage, vc (∞), that can
be derived from (5.56), to Vinc , and letting V = PDIVDC . Table 5.8 shows the PDIVDC
calculated with (5.66). In opposition to the results of Table 5.7, the PDIVDC has a
decreasing trend with temperature and a large sensitivity to it: these two characteristics
were already observed both theoretically and experimentally in [52].
Table 5.8: PDIVDC calculation results using equation (5.66).

T [◦ C] PDIVDC [kV]
20 1468
60 58.1
90 8.91

Figure 5.28 shows the first 12 hours of the cavity voltage transient, vc (t), at the three
different temperatures, simulated using the polymer model (pl) and the RC model (rc)
with an applied voltage of V = 1 kV. The transient at 90 ◦ C reaches the steady state in
both models within the first 2 hours. The same occurs with the voltages at 20 ◦ C and
60 ◦ C simulated using the RC model. The corresponding simulations with the polymer
model do not reach the steady state not even within the first 12 hours. In particular, the

Figure 5.28: Time evolution of vc (t) in the first 12 hours, simulated at the three different temperatures
using the polymer model (pl) and the RC model (rc).

100
 5.5. Internal Partial Discharge

curve 20 ◦ C (pl) is very far away from the steady state. Table 5.9 shows an estimate of
the expected transient duration at different temperatures. Voltage vc (t) was supposed
to be at steady state if vc (t)/vc (∞) < 1.01. In other words, the voltage was assumed
stationary when it was within 1% from the theoretical steady-state voltage vc (∞). In
the case of the RC model (rc) the transient is an exponential, thus it is possible to define
a time constant, τ , and the transient duration can be estimated as t∞ = 5τ . In the case of
the polymer model the estimate was done numerically and two different voltages were
tested: 100 V and 1 kV. At 90 ◦ C the transient duration is similar in all simulations.
There is a slight decrease when the voltage is increased from 100 V to 1 kV in the
polymer model. At 20 ◦ C and 60 ◦ C however, the expected transient duration from the
polymer model is several order of magnitudes higher than it is expected from the RC
model. By all means, the transient duration estimated from the polymer model can be
considered infinite (106 hours correspond approximately to 114 years).
Table 5.9: Estimate of the time required to reach steady state, t∞ , using the RC model (rc) and the
polymer model (pl). The time constant, τ , and the transient duration, t∞ , are expressed in hours.

T [◦ C] τ (rc) t∞ = 5τ (rc) t∞ (pl), 100 V t∞ (pl), 1 kV

20 0.31 1.5 ∼ 1051 ∼ 1051
60 0.31 1.5 ∼ 106 ∼ 106
90 0.26 1.3 1.7 1.4

The transient duration estimated from the polymer model at 20 ◦ C and 60 ◦ C is prob-
ably not precise. Nevertheless, it provides a clear insight on how much the expected
transient duration can differ by using a more refined model (the polymer model) with
respect to the traditional one (the RC model). The study of the transient duration is
extremely important because, under DC excitation, it impacts on the PD repetition rate
and, hence, on the ageing of the insulating material [53].

101
 CHAPTER 6
Conclusions

The present work had the objective of developing dynamic mathematical and physical
models to describe the electric field transient in two main classes of insulating materials
used for HVDC applications, namely oil-paper insulation and polymeric insulation.
Chapter 2 investigated the electric field transient in oil-paper insulation. Starting
from the classical theory of dielectrics and from the Occhini equivalent circuit (also
known as the extended Debye model), the chapter showed that polarization processes
are indeed the dominant process in oil-paper that rule the field transient [15]. A finite
difference scheme was developed to study the field transient in an HVDC cable with
oil-paper insulation. It was demonstrated that the scheme is equivalent to the approach
developed by Occhini [23], who stacked a number of equivalent circuits in series to
approximate the insulation structure [16]. The advantage of the finite difference scheme
is that it can be easily generalized to 2D and 3D geometries, that allow the numerical
investigation of non-symmetric configurations like a defect. The application of the
numerical scheme to a typical 300 kV DC cable showed that polarization processes
slow down the electric field transient in the insulation. Since the maximum field is
reached at DC steady-state, this means that during transient the cable works at lower
stress levels, which is a beneficial effect.
Chapter 3 showed that the Occhini equivalent circuit does not allow to model the
transient observed in polymeric insulation [17]. The logical interpretation of this fact
was that polarization processes may not be the dominant ones in polymeric insulation.
Chapter 4 started from the previous conclusion to develop a model for polymeric
insulation on a different physical basis. The band theory of solids was used to de-
rive a simplified model with non-interacting shallow and deep traps [18]. Some of the
charge carriers (only electrons were considered) would travel in the polymer by hop-
ping through shallow traps, under the driving force of the applied electric field. This

103
Chapter 6. Conclusions

dynamic was empirically described using a power law, based on the experimental be-
haviour observed in PP and XLPE samples. Other carriers (still electrons) would take
part in the dynamic of deep traps, which was observed to be strongly asymmetric: the
deep traps are immediately filled as the field is applied to the sample, while the release
time during the discharge transient is much longer. In this case a more refined model
derived from the works of Shockley and Read [19] was developed. The comparison of
the model simulations with experimental measurements of the PDC currents in PP and
XLPE samples confirmed its validity.
Chapter 5 presented the ks approach to partial discharge modelling. This approach
differs from the well-known Niemeyer approach for one fundamental assumption: it
postulates that, in case of non-uniform field distribution, the field profile can be ap-
proximated by a rectangular profile centred around the peak. The approach had already
been validated by the author of the present work in the case of AC voltage [11]. The
chapter presented two case studies. The first one was that of a surface discharge, which
demonstrates that the ks approach and Niemeyer’s approach yield compatible results
for the PDIV, which confirms the validity of the former. The second case study ap-
plied the dynamic model of polymeric insulation to the calculation of the PDIV for an
internal cylindrical cavity. The results were compared with the traditional RC model
adopted in literature [9, 51]. At DC steady state the two models yield the same results.
During transients instead, the predictions are different, since the polymeric insulation
model has to account for the charging dynamic of the deep traps. The most striking
difference, though, was in the estimation of the transient duration. The RC model pre-
dicted a transient duration of less than 2 hours at all temperatures, namely 20 ◦ C, 60 ◦ C
and 90 ◦ C. The polymeric insulation model, however, predicted a much longer transient
at 20 ◦ C and 60 ◦ C, that was by all practical means infinite. These huge differences in
the prediction of the transient duration may have a significant impact on the estimation
of the electric field that is than used as input in life models to evaluate the life duration
of an insulation system.

104
 Bibliography

[1] G. Mazzanti and M. Marzinotto, Extruded cables for high-voltage direct-current transmission. John Wiley &
Sons, Inc., Hoboken, New Jersey, 1953.
[2] T. W. Dakin, “Electrical insulation deterioration treated as a chemical rate phenomenon,” Transactions of the
American Institute of Electrical Engineers, vol. 67, no. 1, pp. 113–122, 1948.
[3] S. Glasstone, K. J. Laidler, and H. Eyring, The theory of rate processes. McGraw Hill, New York, 1941.
[4] J.-P. Crine, “A molecular model to evaluate the impact of ageing on space charges in polymer dielectrics,”
IEEE Transactions on Dielectrics and Electrical Insulation, vol. 4, no. 5, pp. 487–495, 1997.
[5] G. C. Montanari, “Bringing an insulation to failure: the role of space charge,” IEEE Transactions on Dielectrics
and Electrical Insulation, vol. 18, no. 2, pp. 339–364, 2011.
[6] J. P. Jones, J. P. Llewellyn, and T. J. Lewis, “The contribution of field-induced morphological change to the
electrical aging and breakdown of polyethylene,” IEEE Transactions on Dielectrics and Electrical Insulation,
vol. 12, no. 5, pp. 951–966, 2005.
[7] G. Mazzanti, G. C. Montanari, and L. A. Dissado, “Electrical aging and life models: the role of space charge,”
IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, no. 5, pp. 876–890, 2005.
[8] L. Dissado, G. Mazzanti, and G. C. Montanari, “The incorporation of space charge degradation in the life
model for electrical insulating materials,” IEEE Transactions on Dielectrics and Electrical Insulation, vol. 2,
no. 5, pp. 1147–1158, 1996.
[9] P. H. F. Morshuis and J. J. Smit, “Partial discharges at DC voltage: their mechanism, detection and analysis,”
IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, no. 2, pp. 328–340, 2005.
[10] L. Niemeyer, “A generalized approach to partial discharge modeling,” IEEE Transactions on Dielectrics and
Electrical Insulation, vol. 2, no. 4, pp. 510–528, 1995.
[11] P. Cambareri and G. C. Montanari, “A surface discharge model for the design of surface components of insula-
tion systems under AC stress in industrial electronics environments,” IEEE Journal of Emerging and Selected
Topics in Industrial Electronics, pp. 1–9, 2022.
[12] G. C. Montanari, P. Seri, S. F. Bononi, and M. Albertini, “Partial discharge behavior and accelerated aging
upon repetitive DC cable energization and voltage supply polarity inversion,” IEEE Transactions on Power
Delivery, vol. 36, no. 2, pp. 578–586, 2021.
[13] R. Ghosh, P. Seri, and G. C. Montanari, “Condition assessment of electrical equipment in harsh electrical envi-
ronment,” pp. 1–6, 2019 IEEE 4th International Conference on Condition Assessment Techniques in Electrical
Systems (CATCON), 2019.
[14] A. K. Jonscher, Dielectric relaxation in solids. Chelsea Dielectric Press Ltd London, 1983.
[15] P. Cambareri, C. de Falco, L. Di Rienzo, P. Seri, and G. C. Montanari, “Charging and discharging current mea-
surements and impact of polarization dynamics on electric field modeling in HVDC paper-insulated cables,”
IEEE Access, vol. 9, pp. 45155–45162, 2021.

105
Bibliography

[16] P. Cambareri, L. Di Rienzo, C. de Falco, P. Seri, and G. C. Montanari, “Electric field calculation during voltage
transients in HVDC cables: contribution of polarization processes,” IEEE Transactions on Power Delivery,
vol. 37, no. 6, pp. 5245–5432, 2022.
[17] P. Cambareri, C. de Falco, L. Di Rienzo, P. Seri, and G. C. Montanari, “Simulation and modelling of transient
electric fields in HVDC insulation systems based on polarization current measurements,” Energies, 2021.
[18] P. Cambareri, C. de Falco, L. Di Rienzo, P. Seri, and G. C. Montanari, “Circuital modelling of polarization
and depolarization currents in polymeric materials under low electric fields,” IEEE Transactions on Dielectrics
and Electrical Insulation, 2023.
[19] W. Shockley and W. T. Read, Jr., “Statistics of the recombination of holes and electrons,” Physical Review,
vol. 87, no. 5, pp. 835–842, 1952.
[20] J. G. Simmons and G. W. Taylor, “Nonequilibrium steady-state statistics and associated effects for insulators
and semiconductors containing an arbitrary distribution of traps,” Physical Review B, vol. 4, no. 2, pp. 502–511,
1971.
[21] J. G. Simmons and G. W. Taylor, “High-field isothermal currents and thermally stimulated currents in insulators
having discrete trapping levels,” Physical Review B, vol. 5, no. 4, pp. 1619–1629, 1972.
[22] J. G. Simmons and M. C. Tam, “Theory of isothermal currents and the direct determination of trap parameters
in semiconductors and insulators containing arbitrary trap distributions,” Physical Review B, vol. 7, no. 3,
pp. 3706–3713, 1973.
[23] E. Occhini and G. Maschio, “Electrical characteristics of oil-impregnated paper as insulation for hv dc cables,”
IEEE Transactions on Power Apparatus and Systems, vol. PAS-86, no. 3, pp. 312–326, 1967.
[24] M. J. P. Jeroense and F. H. Kreuger, “Electrical conduction in HVDC mass-impregnated paper cable,” IEEE
Transactions on Dielectrics and Electrical Insulation, vol. 2, no. 5, pp. 718–723, 1995.
[25] M. J. P. Jeroense and H. F. Morshuis, “Electric fields in HVDC paper-insulated cables,” IEEE Transactions on
Dielectrics and Electrical Insulation, vol. 5, no. 2, pp. 225–236, 1998.
[26] P. Debye, Polar molecules. The Chemical Catalogue Company, Inc., New York, 1929.
[27] V. der Houhanessian, Measurement and analysis of dielectric response in oil-paper insulation systems. Phd
dissertation, ETH Zürich, Zürich, Switzerland, 1998.
[28] V. der Houhanessian and W. S. Zaengl, “Time domain measurements of dielectric response in oil-paper insu-
lation system,” vol. 1, (Montreal, Quebec, Canada), pp. 47–52, IEEE International Symposium on Electrical
Insulation, 1996.
[29] V. der Houhanessian and W. S. Zaengl, “On-site diagnosis of power transformers by means of relaxation
current measurements,” vol. 1, (Arlington, VA, USA), pp. 28–34, IEEE International Symposium on Electrical
Insulation, 1998.
[30] G. Hernandez and A. Ramirez, “Dielectric response model for transformer insulation using frequency domain
spectroscopy and vector fitting,” Energies, vol. 15, no. 7, 2022.
[31] F. Yang, L. Du, L. Yang, C. Wei, Y. Wang, L. Ran, and P. He, “A parameterization approach for the dielectric
response model of oil paper insulation using FDS measurements,” Energies, vol. 11, no. 3, 2018.
[32] J. Wei, G. Zhang, H. Xu, H. Peng, S. Wang, and M. Dong, “Novel characteristic parameters for oil-paper
insulation assessment from differential time-domain spectroscopy based on polarization and depolarization
current measurement,” IEEE Transactions on Dielectrics and Electrical Insulation, vol. 18, no. 6, pp. 1918–
1928, 2011.
[33] R. Wiengarten, F. Jenau, and B. Bakija, “Polarization and depolarization behavior of various oil impreganted
paper insulations stressed by very high DC field strength,” (Hannover, Germany), XVII International Sympo-
sium on High Voltage Engineering, 2011.
[34] G. Hakonseth and E. Ilstad, “Time-dependent electric field distribution in layered paper-oil insulation,” No. 26,
(Tampere, Finland), 26th Nordic Insulation Symposium, 2019.
[35] Y. Shi, M. Zhu, Y. Chen, and X. Cai, “Power flow reversal control of hybrid HVDC transmission system with
LCC and FBMMC,” (Osaka, Japan), pp. 1295–1300, 2020 IEEE Region 10 Cconference (TENCON), 2020.
[36] A. K. Jonscher and M. A. Careem, “Non-linearity of dielectric behaviour as evidence of space charge,” Physics
Letters A, vol. 55, no. 4, pp. 257–259, 1975.
[37] L. A. Dissado and R. M. Hill, “Non-exponential decay in dielectrics and dynamics of correlated systems,”
Nature, vol. 279, no. 5715, pp. 685–689, 1979.

106
 Bibliography

[38] R. A. Smith, Semiconductors. Cambridge University Press, Cambridge, 1959.

[39] F. Baudoin, S. Le Roy, G. Teyssedre, and C. Laurent, “Bipolar charge transport model with trapping and
recombination: an analysis of the current versus applied electric field characteristic in steady state conditions,”
Journal of Physics D: Applied Physics, vol. 41, no. 025306, 2008.
[40] S. Le Roy, G. Teyssedre, C. Laurent, G. C. Montanari, and F. Palmieri, “Description of charge transport in
polyethylene using a fluid model with a constant mobility: fitting model and experiments,” Journal of Physics
D: Applied Physics, vol. 39, pp. 1427–1436, 2006.
[41] G. Mazzanti, G. C. Montanari, and J. M. Alison, “A space-charge based method for the estimation of apparent
mobility and trap depth as markers for insulation degradation - theoretical basis and experimental validation,”
IEEE Transactions on Dielectrics and Electrical Insulation, vol. 10, no. 2, pp. 187–197, 2003.
[42] X. He, P. Seri, I. Rytoluoto, R. Anyskza, M. Niittymaki, A. Mahtabani, H. Naderiallaf, E. Saarimaki, K. Lahti,
M. Paajanen, W. Dierkes, and A. Blume, “Influence of polar and unpolar silica functionalization on the di-
electric properties of PP/POE nanocomposites,” (Valencia, Spain), pp. 229–232, 2020 IEEE 3rd International
Conference on Dielectrics, 2020.
[43] Y. Zheng, H. Huang, X. Zhong, and Y. V. Serdyuk, “Simulation of charge transport in polypropylene-based
nano-composites,” Journal of Physics D: Applied Physics, vol. 54, no. 235501, 2021.
[44] J. Xue, Z. Zhang, M. Zhu, Y. Zhao, and L. Ding, “Charge transport behavior in xlpe-insulated cables under DC
and thermal gradient coupling field considering the effects of non-uniform conductivity,” Journal of Physics
D: Applied Physics, vol. 54, no. 485501, 2021.
[45] P. K. Watson, “The transport and trapping of electrons in polymers,” IEEE Transactions on Dielectrics and
Electrical Insulation, vol. 2, no. 5, pp. 915–924, 1995.
[46] G. C. Montanari, P. Seri, M. Ritamäki, K. Lahti, I. Rytöluoto, and M. Paajanen, “Performance of nanoparticles
in the electrical behavior of DC capacitor films,” (Xi’an, China), pp. 41–44, 12th International Conference on
the Properties and Applications of Dielectric Materials (ICPADM), 2018.
[47] J. M. Meek and J. D. Craggs, Electrical breakdown of gases. Oxford at the Clarendon Press, 1953.
[48] S. Nijdam, J. Teunissen, and U. Ebert, “The physics of streamer discharge phenomena,” Plasma Sources
Science and and Technology, vol. 29, no. 103001, 2020.
[49] L. Niemeyer, L. Ullrich, and N. Wiegart, “The mechanism of leader breakdown in electronegative gases,” IEEE
Transactions on Dielectrics and Electrical Insulation, vol. 24, no. 2, pp. 309–324, 1989.
[50] E. Seran, M. Godefroy, E. Pili, N. Michielsen, and S. Bondiguel, “What we can learn from measurements of
air electric conductivity in 222 Rn-rich atmosphere,” Earth and space science, vol. 4, pp. 91–106, 2017.
[51] P. Seri, H. Naderiallaf, and G. C. Montanari, “Modelling of supply voltage frequency effect on partial discharge
repetition rate and charge amplitude from AC to DC at room temperature,” IEEE Transactions on Dielectrics
and Electrical Insulation, vol. 27, no. 3, pp. 764–772, 2020.
[52] P. Seri, L. Cirioni, H. Naderiallaf, G. C. Montanari, R. Hebner, A. Gattozzi, and X. Feng, “Partial discharge
inception voltage in DC insulation systems: a comparison with AC voltage supply,” pp. 179–179, 2019 IEEE
Electrical Insulation Conference (EIC), 2019.
[53] H. Naderiallaf, P. Seri, and G. C. Montanari, “Effect of voltage slew rate on partial discharge phenomenology
during voltage transient in HVDC insulation: the case of polymeric cables,” IEEE Transactions on Dielectrics
and Electrical Insulation, vol. 36, no. 2, pp. 578–586, 2021.

107