Analysis of Dielectric Response

Analysis of Dielectric Response Measurement Methods and Dielectric Properties of Resin-Rich Insulation During Processing
by Anders Helgeson
Submitted in partial fulfilment of the requirements for the degree of Doctor of Technology
TRITA EEA-0002 ISSN 1100-1593
Kungl Tekniska Hgskolan Department of Electric Power Engineering Division Electrotechnical Design Stockholm, Sweden
Abstract
Abstract
The dielectric diagnostic methods of polarisation/depolarisation currents and recovery voltage, which are time domain methods, and capacitance and tan measurements at different frequencies, which is a frequency domain method, have been studied here. In the first part of the thesis, these measurement methods are analysed and evaluated. Three different types of insulation systems are included in the evaluation, oil/paper (high loss), resin-rich mica tape (medium loss) and cross-linked polyethylene, XLPE, (low loss). These three systems have different dielectric response in both shape and magnitude. Conclusions are made regarding choice of measurement method depending on the dielectric response of the insulation material. Examples are also given of how to estimate conductivity and the dielectric response function in the time domain from measurements with a finite charging period. Furthermore, relations between time domain and frequency domain and the possibilities of Fourier transforming data from one domain to the other are discussed. Fourier transforming is done with a spline approximation technique, the Hamon approximation and by fitting base functions that have analytic Fourier transforms to measured data. These techniques are reviewed with special attention to the problem of estimating data outside the measurement window. The second part of the thesis includes studies of the change of dielectric properties during the manufacturing stage of a composite insulation used in high voltage rotating machines. The insulation consists of a resin-rich mica tape with woven glass or polyester film (PET) as carrier material. The aim was to improve the quality of the composite insulation by optimising the heat and pressure cycle used in the production. A test cell has been designed and built to be able to process simple parallel plate samples under conditions similar to the factory process. With a maximum heating rate of 9C/min and a maximum cooling rate of 20C/min arbitrary temperature paths could be programmed. All samples were processed under static pressure in the MPa range. The chemical reaction during curing of the resin-rich mica tape was studied with differential scanning calorimetry (DSC). A simple reaction rate model was fit to the DSC measurements making it possible to calculate the degree of curing during an arbitrary temperature path. The change of dielectric response with time during curing under different temperature paths and at a constant pressure was monitored. Both laboratory experiments and factory measurements have been made and based on these measurements a simple network model is proposed to explain the measured dielectric response in terms of material structure and degree of curing.
Acknowledgements
Acknowledgements
During the last five years, I have had the pleasure to work within the Electrical Insulation group at the Department of Power Engineering at Kungl. Tekniska Hgskolan (KTH) in Stockholm, Sweden. This work resulted in my Ph.D. thesis, which is presented here. The project has been financially supported by ABB within the ELEKTRA program. I am very grateful to my supervisor Prof. Uno Gfvert for his guidance, never-ending enthusiasm and very inspiring discussions (no coffee break on Thursdays). This work would not have been possible without him. I would also like to thank the head of the department Prof. Roland Eriksson, who initiated this project and who has always been very positive and supportive throughout these years. Throughout this work, I have had the great opportunity to work together with Dr. Ron Brammer, Mr. Dick Rudolfsson, Mr. Stefan Engelbertsson and Mr. Bo Ullberg at ABBAlstom Power in Vsters, Sweden. I especially want to thank Dr. Ron Brammer and Mr. Dick Rudolfsson for their invaluable help during the preparation of this thesis. All the DSC measurements were made at the department of Polymer Technology at KTH and I am thankful to Dr. Patrik Roseen, Mr. Henrik Hillborg and Mr. Kamyar Fateh-Alavi for assisting me. The mechanical tests were made at ABB Corporate Research in Vsters, Sweden and I would like to thank Mr. Jarmo Khknen for helping me. This work has been carried out within the Electrical Insulation group. I am very grateful to present and former members of this group: Prof. Stanislaw Gubanski, Mr. Peter Werelius, Tech. Lic. Bjrn Holmgren, Tech. Lic. Peter Thrning, Tech. Lic. Roberts Neimanis, Mr. Hans Edin and Tech. Lic. Mats Kvarngren. I would like to thank the rest of the staff at the department for these years, especially present and former member of the lunch and coffee break team: Dr. Niklas Magnusson, Dr. Fredrik Stillesj, Tech. Lic. Pr Holmberg, Dr. Anders Lundgren, Dr. Anders Bergqvist, Tech. Lic. Niclas Schnborg and Ms. Anna Wolfbrandt. I would also like to thank Mrs. Marianne Ahlns, Mr. Gte Bergh and for technical assistance when building the test cell Mr. Yngve Eriksson and Mr. Olle Brnvall. Many greetings to Roebels bar! Finally, I want to express my deepest gratitude and love to my mother, father and girlfriend, this work would never have been possible without their support and patience. Things are definitely not practical but very interesting and giving. Yes I know, you have traffic lights!
Anders Helgeson Stockholm April 28, 2000
Contents
Contents
1
1.1 1.2 1.3
Introduction_________________________________________ 1
Background ................................................................................................................... 1 Summary of the thesis ................................................................................................... 2 Publications ................................................................................................................... 4
Review of electrical insulation systems in high-voltage rotating machines ____________________________________ 5

2.1 Introduction ................................................................................................................... 5 2.2 Different groundwall insulation systems....................................................................... 5 2.3 Manufacturing process of stator coils ........................................................................... 6 2.4 Semi-conducting tapes, slot fillers and stress grading for stator coils .......................... 7 2.5 Degrading mechanisms for insulating materials ........................................................... 8 2.6 Today existing diagnostic methods for high-voltage rotating machines....................... 8 2.6.1 Off-line diagnostic methods .................................................................................... 8 2.6.2 On-line diagnostic methods ..................................................................................... 9
Electric properties of dielectric materials _______________ 11

3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 Introduction ................................................................................................................. 11 Electrostatics of conductors......................................................................................... 11 Electrostatics of dielectrics.......................................................................................... 12 Electric polarisation mechanisms................................................................................ 14 Dielectric response in time domain ............................................................................. 17 Dielectric response in frequency domain .................................................................... 18 Time-dependence of dielectrics................................................................................... 20 Kramers-Kronig relations............................................................................................ 24 Heterogeneous dielectrics............................................................................................ 27 Temperature dependence of dielectrics ....................................................................... 29
Dielectric measurement techniques ____________________ 33

4.1 Introduction ................................................................................................................. 33 4.2 Polarisation and depolarisation currents...................................................................... 34 4.2.1 Estimation of the dielectric response function....................................................... 37 4.2.2 Estimation of the conductivity............................................................................... 39 4.3 Recovery voltage......................................................................................................... 40 4.3.1 The forward problem ............................................................................................. 41 4.3.1.1 Equations giving the analytic expression for the recovery voltage................ 42 4.3.1.2 Numerical calculation of the recovery voltage............................................... 43 4.3.1.3 Recovery voltage measurements .................................................................... 44 4.3.2 The inverse problem .............................................................................................. 47 4.3.2.1 Constrained minimisation of the current ........................................................ 47 4.3.3 Memory effects ...................................................................................................... 51 4.4 Loss factor and capacitance......................................................................................... 52 4.5 How to choose the best measurement method ............................................................ 55 4.6 Relations between time and frequency domain........................................................... 56 4.6.1 The Hamon approximation .................................................................................... 58 iii
Contents 4.6.2 Numerical Fourier transform ................................................................................. 60 4.6.2.1 Transformation from time to frequency domain ............................................ 62 4.6.2.2 Transformation from frequency to time domain ............................................ 64 4.6.3 Fourier transform using analytic base functions.................................................... 66 4.6.4 Numerical calculation of the Kramers-Kronig relations ....................................... 70 4.7 Summary ..................................................................................................................... 75
Test cell for processing resin-rich insulation _____________ 77

5.1 Introduction ................................................................................................................. 77 5.2 Designing the test cell ................................................................................................. 77 5.2.1 Dimensioning heating............................................................................................ 79 5.2.2 Dimensioning cooling............................................................................................ 81 5.3 Controlling the temperature......................................................................................... 83 5.4 Controlling the pressure .............................................................................................. 86 5.5 Measurement circuit .................................................................................................... 88
Measurements on resin-rich insulation _________________ 93

6.1 Introduction ................................................................................................................. 93 6.2 Preparing test samples for a dielectric response measurement ................................... 93 6.3 Preparing test samples for a differential scanning calorimetry (DSC) measurement . 95 6.4 Woven glass as carrier material................................................................................... 96 6.4.1 The uncured system ............................................................................................... 97 6.4.2 The fully cured system........................................................................................... 98 6.4.3 Different degrees of curing .................................................................................... 99 6.4.3.1 Degree of curing, 0.4 ............................................................................. 100 6.4.3.2 Degree of curing, 0.6 ............................................................................. 101 6.4.3.3 Degree of curing, 0.8 ............................................................................. 102 6.4.3.4 Degree of curing, 0.9 ............................................................................. 103 6.4.3.5 Loss, tan, plotted against degree of curing .............................................. 104 6.4.4 Dielectric response along the surface of the resin-rich mica tape ....................... 105 6.4.4.1 The uncured system...................................................................................... 106 6.4.4.2 The fully cured system ................................................................................. 107 6.4.5 Curing under different temperature paths............................................................ 108 6.4.5.1 Curing under constant temperature .............................................................. 109 6.4.5.2 Curing under a temperature ramp................................................................. 111 6.4.5.3 Curing under Factory temperature path .................................................... 113 6.4.5.4 Curing under Soft temperature path.......................................................... 115 6.4.5.5 Curing under Hard temperature path ........................................................ 118 6.4.6 Mechanical measurements................................................................................... 120 6.4.7 Factory measurement during pressing of stator bar............................................. 122 6.4.7.1 Temperature measurement ........................................................................... 123 6.4.7.2 Capacitance and tan measurement.............................................................. 125 6.4.7.3 Tip-up measurement after pressing the test stator bar.................................. 127 6.4.7.4 Summary ...................................................................................................... 128 6.5 PET-film as carrier material ...................................................................................... 128 6.5.1 The uncured system ............................................................................................. 129 6.5.2 The fully cured system......................................................................................... 130 6.5.3 Different degrees of curing .................................................................................. 131 iv
Contents 6.5.3.1 Degree of curing, 0.1 ............................................................................. 132 6.5.3.2 Degree of curing, 0.3 ............................................................................. 133 6.5.3.3 Degree of curing, 0.5 ............................................................................. 134 6.5.3.4 Degree of curing, 0.8 ............................................................................. 134 6.6 Summary ................................................................................................................... 135
Modelling a resin-rich insulation system _______________ 137

7.1 Introduction ............................................................................................................... 137 7.2 Modelling the curing of a resin ................................................................................. 137 7.2.1 Resin-rich mica tape with woven-glass as carrier material ................................. 139 7.2.1.1 Different degrees of curing........................................................................... 141 7.2.1.2 Curing under constant temperature .............................................................. 142 7.2.1.3 Curing under temperature ramp ................................................................... 143 7.2.1.4 Curing under Factory, Soft and Hard temperature paths ................... 144 7.2.2 Resin-rich mica tape with polyester film as carrier material ............................... 145 7.2.2.1 Different degrees of curing........................................................................... 146 7.3 Temperature dependence for the resin-rich mica tapes............................................. 147 7.3.1 The resin rich-mica tape with woven glass as carrier material............................ 149 7.3.1.1 Degree of curing, 0 ................................................................................ 149 7.3.1.2 Degree of curing, 0.4 ............................................................................. 150 7.3.1.3 Degree of curing, 0.6 ............................................................................. 151 7.3.1.4 Degree of curing, 0.8 ............................................................................. 151 7.3.1.5 Degree of curing, 0.9 ............................................................................. 152 7.3.1.6 Degree of curing, 1 ................................................................................ 153 7.3.1.7 Comparison between different degrees of curing......................................... 153 7.3.1.8 Comparison between fully cured samples cured under Factory, Soft and Hard temperature paths ............................................................................. 155 7.3.2 The resin rich-mica tape with polyester film (PET) as carrier material .............. 157 7.3.2.1 Degree of curing, 0 ................................................................................ 157 7.3.2.2 Degree of curing, 0.1 ............................................................................. 158 7.3.2.3 Degree of curing, 0.3 ............................................................................. 159 7.3.2.4 Degree of curing, 0.5 ............................................................................. 159 7.3.2.5 Degree of curing, 0.8 ............................................................................. 160 7.3.2.6 Degree of curing, 1 ................................................................................ 161 7.3.2.7 Comparison between different degrees of curing......................................... 161 7.4 Modelling of the dielectric response from resin-rich mica tapes .............................. 163 7.4.1 Network model .................................................................................................... 163 7.4.2 Dielectric response of the resin-rich mica tape with woven glass as carrier material ................................................................................................................ 165 7.4.2.1 Assumption epoxy-resin: dipole + conductivity .......................................... 167 7.4.2.2 Assumption epoxy-resin: LFD + conductivity............................................. 169 7.4.3 Dielectric response of the resin-rich mica tape with PET-film as carrier material ............................................................................................................................. 171 7.5 Summary ................................................................................................................... 174
8
8.1 8.2
Conclusions _______________________________________ 177

Analysis of dielectric measurement methods............................................................ 177 Study of resin-rich mica tape insulation.................................................................... 178 v
Contents
Future work_______________________________________ 181
10 Bibliography ______________________________________ 183 Appendix A: Orientational polarisation ___________________ 191 Appendix B: Analytic solution to the Kramers-Kronig relations for the Curie-von Schweidler model___________ 193 Appendix C: Numerical Fourier transform by approximating data with cubic splines ______________________ 197 Appendix D: The time domain solution to the network model _ 201 Appendix E: List of symbols _____________________________ 205 Appendix F: Pictures of the test cell ______________________ 209
vi
Introduction
Chapter 1
Introduction
1.1
Background
This Ph.D. thesis is the result of a project, which started in the beginning of 1995 at the department of Power Engineering at Kungl. Tekniska Hgskolan in Stockholm, Sweden. The idea was to study fundamental dielectric properties of insulation in high-voltage rotating machines with diagnostic methods developed at the department. The existing methods at that time were in the frequency domain, a high voltage measurement system for capacitance and tan at variable frequency [98] and in the time domain, a high voltage polarisation and depolarisation current and recovery voltage measurement system [45]. It was of interest to understand these measurement methods, their practical advantages and disadvantages especially regarding diagnostics of rotating machines. An aim was to link this understanding to existing diagnostic methods, which are in use today. The idea from the beginning was to build up a new measurement system mainly to incorporate the time and frequency domain systems into one system but also to build up knowledge in sensitive low voltage and low current measurement techniques [37]. In parallel with this, it was natural to study techniques to extract material characterising parameters from the measurement data and to transform data between domains. From the very beginning, the idea was to test the diagnostic methods by measuring insulation systems with different dielectric response in both shape and magnitude. The motivation for studying rotating machines in particular is that the electrical insulation is a very important part in the whole machine. The condition of the insulation will determine both economical and technical lifetime as well as quality of the power generated or distributed. Different types of pre-impregnated (resin-rich) mica tapes are commonly used as groundwall insulation. Their final mechanical, thermal and electrical properties are closely related to the manufacturing stage and especially the curing process of the epoxy resin. For example, if internal stresses are built in during curing this can result in poor binding between layers or to the insulated conductor. When such problems occur in a coil, it will most likely not pass factory tests or it will have a reduced service life. The first step in studying the manufacturing step was to scale down the factory process to an easy to handle test cell in the laboratory. A test cell was constructed and built in the workshop and could press parallel plate samples under constant pressure and a pre-programmed temperature path. The first specifications for the test cell were to match heating and cooling rates as found in the factory process. Several rather complex steps involving cycling the pressure were skipped and after discussions it was decided that static pressure would be enough as long it was so high that the resin was evenly distributed between each insulation layer and that possible excess resin between layers was pressed away.
Chapter 2
Review of electrical insulation systems in high-voltage rotating machines
Two questions were raised from the beginning. What would the change in dielectric response look like during curing and was it possible to read out a specific degree of curing from a dielectric response measurement. A rather simple idea was stated of how the rate of crosslinking in the epoxy-resin would affect the quality of the insulation in terms of built in tension and adhesion between layers and towards the conductor. A high rate of curing would quickly freeze the system leaving out possible cross-links whereas a slower rate would ensure that all cross-links would find each other. However, one limiting economical factor in the production of stator coils turned out to be time and especially time in the press. So, was it possible to find an optimal curing process from both an economical and insulation quality point of view?
1.2
Summary of the thesis
This chapter contains a brief summary of the contents of this thesis. The summary is made in order to focus on the contributions, which are made. Chapter 1 gives the background to this project and a summary of what is found in this Ph.D. thesis, followed by a list of publications. In chapter 2, a short review discusses common features concerning insulation systems and diagnostic methods for rotating machines. Many factors interact and contribute to the ageing of the groundwall insulation material. Important decisions concerning the condition of the complete machine are often made based on the results from the diagnostic methods. It is important for a user of a generator or a motor to know and understand what can go wrong in the insulation and how this can affect the properties of the generator or the motor. One of the ultimate goals with diagnostics is to estimate the residual life of the insulation. Diagnostics of generators or motors are especially interesting since a failure in either may cost a lot of money in loss of production. In chapter 3, basic concepts of dielectric materials are discussed. A brief introduction is made to electric polarisation in a static electric field and to different polarisation mechanisms. Models for heterogeneous materials are discussed to raise a few questions how to model the composite insulation in rotating machines. This chapter continues by showing that the electric behaviour of a dielectric material can be described either in the time domain (, , f(t)) or in the frequency domain (, , ()). If the material is linear, the information obtained in either domain will be equivalent. The Kramers-Kronig relations are introduced and the master curve shifting technique is discussed for modelling the temperature dependence. Fundamental relations between three different measurement methods are studied in chapter 4. The three methods are polarisation/depolarisation currents and recovery voltage in the time domain and capacitance/tan in the frequency domain. In this chapter, three different electrical insulation systems will be studied with the three different methods mentioned above at room temperature. The insulation systems are oil/paper insulation, resin-rich mica tape insulation and cross-linked polyethylene (XLPE) insulation. These three insulation systems are commonly used today in high voltage applications and they all have different dielectric response ranging from higher, oil/paper, towards lower loss, XLPE. It will be shown what type of information is realistic to extract not only from each method but also from each insulation system. Conclusions are made about choice of method depending on insulation system response and on the type of information and physical mechanism wanting to be 2
Introduction
Chapter 1
diagnosed. Furthermore, relations between time domain and frequency domain and the possibilities of Fourier transforming data from one domain to the other are discussed. Fourier transforming is done with a spline approximation technique, the Hamon approximation and by fitting base functions that have analytic Fourier transforms to measured data. These techniques are reviewed, especially the problem of estimating data outside the measurement window. In chapter 5, all the design steps of the test cell are described. The basic idea behind building a test cell was to simulate the manufacturing process of the electrical insulation in rotating machines. The insulation system studied here is based on a pre-impregnated (resin-rich) mica tape that has to be cured under a given temperature and pressure profile in order to reach maximum electrical, thermal and mechanical performance. The test cell was built to match these demands. In chapter 6, the dielectric response in the frequency domain has been measured on resin-rich mica tape insulation used as electrical insulation in rotating machines. Two different types of resin-rich mica tapes have been studied here. The first tape has a carrier material made of woven glass and the second one has a polyester film (PET). Low voltage measurements, 5 to 200 V, in the frequency range 10 mHz to 1 MHz were made using the test cell described in chapter 5. The main idea was to use non-destructive dielectric response measurements to study the manufacturing stage of resin-rich mica tape insulation. Simple parallel plate samples were pressed and cured under varying conditions in order to investigate the relation between the measured dielectric response and the degree of curing. The dielectric response was measured before, during, and after the processing cycle. Several different temperature paths were used including processing under constant temperature. Experimental results were compared to a full-scale factory measurement. In order to try verifying the assumption that a slow temperature path compared to a fast gave better mechanical properties a simple mechanical test of the adhesion between the copper conductor and the insulation were performed. The chemical reaction during curing of the two resin-rich mica tapes was also studied with differential scanning calorimetry (DSC). Making a DSC measurement and relating this measurement to the measurement of an uncured sample would determine the degree of curing. In chapter 7, a simple Arrhenius activated reaction model was used to describe the progress of curing, change in , for the two resin-rich mica tapes studied. The parameters in the reaction model were estimated from differential scanning calorimetry (DSC) measurements. When these parameters were known the degree of curing could be calculated for an arbitrary temperature path which made it possible to relate degree of curing, , to the measured dielectric response, capacitance and tan. The master curve approach is used here to study the temperature behaviour, thermal activation, and how this is changing with degree of curing, . It is expected that the thermal activation will be reduced with degree of curing, which relates both to the change in viscosity and the glass transition temperature during the curing process. To explain the dielectric response from the resin-rich mica tapes a circuit network model is proposed consisting of the materials building up the tapes. Mica and the two tape carrier materials woven glass and polyester (PET) are supposed to be unchanged during the curing 3
Chapter 2
process and will only have a temperature dependence. The dielectric response of the resin will change with both degree of curing, , and temperature. Certain features like barrier effects can be explained with the geometrical configuration. In chapter 8, the conclusions from this Ph.D. thesis are made. Suggestions for further studies and future work are mentioned in chapter 9.
1.3
Publications
Throughout the work in this project the following publications have been written 1. A. Helgeson: Calculation of the Dielectric Response Function from Recovery Voltage Measurements, Master Thesis (in Swedish), A-EEA-9408, Dept. Electric Power Engineering, Kungl. Tekniska Hgskolan, Stockholm, Sweden, 1994. 2. U. Gfvert, S. M. Gubanski, A. Helgeson: Calculations of Thermally Stimulated Depolarisation Current from Isothermal Dielectric Response Data, Proc. of the 1995 International Symposium on Electrical Insulating Materials, Tokyo, Japan, 1995. 3. A. Helgeson, U. Gfvert: Calculation of the Dielectric Response Function from Recovery Voltage Measurements, Proc. of the 1995 Conference on Electrical Insulation and Dielectric Phenomena, Virginia Beach, Virginia, USA, 1995. 4. A. Helgeson: Dielectric Properties of Machine Insulation studied with Dielectric Response, Licentiate Thesis, Kungl. Tekniska Hgskolan, TRITA EEA-9704, ISSN 1100-1593, Stockholm, Sweden, 1997. 5. A. Helgeson, U. Gfvert: Dielectric Response Measurements in Time and Frequency Domain on High Voltage Insulation with Different Response, Proc. of the 1998 International Symposium on Electrical Insulating Materials, Toyohashi, Japan, 1998. 6. A. Helgeson, U. Gfvert: Dielectric Response During Curing of a Resin-Rich Insulation System for Rotating Machines, Proc. of the 1999 Conference on Electrical Insulation and Dielectric Phenomena, Austin, Texas, USA, 1999.
Chapter 2
2.1
Introduction
This short review discusses common ideas concerning insulation systems and diagnostic methods for rotating machines [24], [10]. The review will give a background to the diagnostic methods discussed in chapter 4. The most common purpose of insulation diagnostics is to detect ageing [63]. Many factors interact and contribute to the ageing of the groundwall insulation material. Important decisions concerning the condition of the complete machine are often made based on the results from the diagnostic methods. It is important for a user of a generator or a motor to know and understand what can go wrong in the insulation and how this can affect the properties of the generator or the motor [27], [82]. One of the ultimate goals with diagnostics is to estimate the residual life of the insulation. Diagnostics of generators or motors are especially interesting since a failure in either may be expensive due to loss of production.
2.2
Different groundwall insulation systems
Shellac micafolium: Here mica flakes are bound together with shellac to form sheets, which are wrapped and hot pressed around the slot section of the winding. This system contains many voids due to evaporation of volatiles in the shellac during the process stage. The many voids make this system suffer from heavy partial discharge erosion and poor heat transfer. This kind of insulation system was used from the beginning of 1900 to the sixties [76]. Asphalt-bonded mica tape (1.8 kV/mm): The entire coil is insulated with asphalt-bonded mica tape. Asphalt-mica is a thermoplastic insulation system. The asphalt-bonded mica tape system can be vulnerable to tape separation due to thermal cycling (this is especially true for large hydropower generators which have problems with oil vapour). This insulating system was used from the forties to the seventies [76]. Synthetic resin bonded mica tape (2.4 kV/mm): This system consists of small mica flakes that are deposited on glass fibre backing tape see Figure 2.1. Once the tape has been wound on the conductors, the synthetic resin is cured at elevated temperature and pressure. This technique offers the possibility of manufacturing nearly void-free insulation that can withstand high dielectric stress. This insulation system has been in use since the sixties [7], [16], [76], [83]. Silicone Rubber: Silicone rubber is used especially in applications that need to withstand high temperatures. Even though there is no mica in this system, it can withstand partial discharges 5
Chapter 2
(PD) well. Silicone rubber is vulnerable to mechanical damages and this restricts the voltage level to 6 kV. Silicon rubber can be used together with mica to increase the voltage level. However, the costs for such an insulating system are higher than for an epoxy system [76].
Figure 2.1
Resin-rich mica tape with woven glass as carrier material used as groundwall insulation in rotating machines. In the background, different types of strand and turn insulation are shown. (Picture from ABB-Alstom Power AB).
2.3
Manufacturing process of stator coils
Manufacturing of stator coils for high-voltage rotating machines when using mica tapes and synthetic resins can be divided in two different categories. The first alternative is to use a resin-rich mica tape with the epoxy-resin in a B-stage already in the tape and the second is to use a dry mica tape that is first vacuum pressure impregnated (VPI) and then cured. There are advantages and disadvantages with both ways of producing coils. To produce a coil the first step is to form geometrically the coil from the desired number of conductors all individually insulated see Figure 2.2. It is crucial that the right dimensions are obtained and that the conductors bond together. When using resin-rich mica tapes the coil is most often wrapped in half-lap, either by hand or with a taping machine with a tension of about 40-60 N see Figure 2.2. It is important to apply the right number of turns to get the correct insulation thickness so that the coil will fit into the stator core. There is a compression ratio of about 25-30% after pressing. The coil is loaded into a hot press, around 100C, with suitable release films. Two temperature levels are used, the first to soak the whole coil with the resin and the second to cure the insulation this is done at, for example, 170C for 30-40 min. Pressure is applied to ensure that the resin is correctly distributed throughout the coil and that the right geometrical dimensions are reached. A good press procedure is important to avoid internal stresses being built in during curing, which can result in poor binding between layers or to the insulated conductor. It is also important to avoid wrinkles in the tape especially at corners since this could otherwise lead to electric field enhancement. The coils are installed and connected in the stator core after pressing and a post-curing phase will take place, normally for around 12 hours at 140C. During the post-curing phase the coil ends are cured, usually under pressure from shrink tape. 6
Chapter 2
Figure 2.2
Cross section of a statorbar, showing the different layers for insulation, strand, turn and main (groundwall), slot corona protection and end corona protection used as stress grading. (Picture from ABB-Alstom Power AB).
When using the VPI technique the coil is wrapped with a dry mica tape. The tension and amount of overlapping will affect the impregnation process [17], [102], [103]. The mica paper is mostly of uncalcined type, which has a high porosity, making it easier to impregnate. Depending on the size of the impregnation tank a whole stator or for large machines single stator bars will be impregnated. In case of impregnation of a whole stator, the coils are put into the stator core and the whole winding is completed. Then predrying takes place 1-12 hours at 100C after which the stator is lifted into the VPI tank and vacuum (<0.1-1 mbar) is pumped for 1-8 hours. Room temperature is reached before the resin is pumped in and the tank is filled to about 10 cm above the stator, this procedure takes 0.1-2 hours. Pressure of 210 bar presses the resin into the whole winding, this procedure takes around 2-12 hours or more depending on the viscosity of the resin. This is the most crucial stage since the quality of the whole winding depends on its being fully impregnated. After this stage the stator is lifted out of the tank and into an oven and cured at 120C-150C for 8-20 hours depending on the size and which VPI resin is used. Post-curing can be an extra operation or can take place during the first weeks of operation.
2.4
Semi-conducting tapes, slot fillers and stress grading for stator coils
It is common to provide stator coils with semi-conducting tapes see Figure 2.2 and filler materials in the slot portion. The semi-conducting tapes will provide good electrical contact to the slot wall and the filler materials will protect the groundwall insulation against physical damage as well as fill out the space between the coil surface and the slot. Silicon carbide filled coatings in the form of paints or tapes are applied to the ends of the stator coil to provide voltage stress grading which limits erosion from partial discharges. The stress grading has a voltage dependent resistivity [22].
Chapter 2
2.5
Degrading mechanisms for insulating materials
Thermal: As organic materials form a major part of rotating machine insulation, thermal factors influence the ageing. It is common to express the thermal lifetime with the Arrhenius relationship, Life(L)=A*exp(B/T) where T is the temperature and A, B are material constants [76], [91]. Electrical: Electrical degradation can take the form of partial discharge activity [11], [76]: Internal partial discharges in the bulk of the groundwall isolation can occur and can become very serious if the strands of the conductor become loose. High intensity of external partial discharge in the slot can by time erode the groundwall insulation. However, before that the slot corona protection will be deteriorated and harmful ozone levels reached. Partial discharges in endwing areas are often less life threatening, although they can damage surface coatings and produce ozone.
Mechanical: During normal operation the generator or motor stator coils are exposed to electromechanical forces that can result in erosion of the ground insulation if the coil is not firmly anchored in the slot. Under short circuit conditions, very strong electromechanical forces can lead to cracking of the endwinding portion of the stator coils [63], [76]. Ambient: Motors and generators are often exposed to severe environmental conditions such as moisture, oil and dirt. Problems can occur when, for example, oil dissolves bonding materials and leads to the development of cracks in the groundwall insulation. A common problem is the combination of oil and dirt, which can block ventilation ducts and can cause overheating [76]. Duty: The mode of operation has an important influence on the service life of the generator or motor. Frequent load cycling can contribute to delaminating of the ground insulation resulting in increased internal partial discharge activity [76].
2.6
Today existing diagnostic methods for high-voltage rotating machines
Different diagnostic methods that are in use today will be reviewed here to get a wider perspective and to understand the importance and complexity of the electrical insulation in a rotating machine. A few of these methods like return voltage, insulation resistance, polarisation index and dissipation factor measurements are closely related to methods discussed later in this thesis. Having this in mind when reading those parts of the thesis might introduce some new ideas although this is not the main focus of the thesis.
2.6.1
Off-line diagnostic methods
Visual Inspection: Visual inspection of the end windings and the ends of the slots is important because it reveals some types of deterioration. When the field poles or the entire rotor is removed, it is common to check the tightness of the slot wedges [27], [58]. 8
Chapter 2
Diagnostics Based on Operation History: The operating hours of a stator winding have been considered a standard index for insulation life. Start and stop of a generator or motor and frequent load cycling accelerates the ageing of stator insulation [63], [58]. Return Voltage: A DC-voltage is applied over the ground insulation of the generator or motor followed by a short-circuit and finally after opening the short-circuit the return voltage can be measured. This is a very simple method to investigate whether the insulation resistance and the time constants of the insulation system change with ageing [3]. Insulation Resistance: Insulation resistance is in North America defined as the resistance measured by applying a negative DC-voltage for one minute. It is useful to measure the resistance because it can give an indication if the insulation is wet and contaminated. However, partially wet insulation is not detected with this method. To detect such a problem it is necessary to make a dissipation factor measurement (tan). The measured insulation resistance value is strongly temperature dependent [27], [33], [58], [84], [85]. Polarisation Index: The polarisation index is defined as the ratio of the insulation resistance at ten minutes to that at one minute. The polarisation index indicates the same sort of defects as were mentioned for the insulation resistance with the exception that the polarisation index is less temperature dependent [27], [58], [84], [85]. Direct-Voltage Ramp Test: Here the resistance of the ground insulation is measured as a function of a direct voltage ramp. This is a non-destructive method to detect cracks and fissures [27]. Dissipation Factor Tip-Up: Here the loss factor tan as a function of voltage is measured. This should result in a horizontal line since tan is independent of voltage, but when partial discharges occur there is a tip-up in the curve. This gives an indication if there is a likelihood of partial discharges under normal operation of the generator or motor [27], [33], [58], [84], [85]. Electromagnetic Probe: Sometimes it can be difficult to perform a tip-up measurement for a single coil. A way to overcome this problem is to use an electromagnetic probe and a radio noise meter to detect partial discharges associated with a coil in a specific stator slot. A problem here is to find the right frequency to tune the probe to [76].
2.6.2
On-line diagnostic methods
Partial Discharges: Partial discharge measurements can be divided into three parts [62] Partial discharge pulse analysis can be applied in generators to distinguish slot discharges from external discharges on the end parts of the winding [11]. Slot discharges together with coil vibrations are very energetic and will cause damage. Partial discharges can be measured with a movable radio frequency antenna. With the help of the antenna, it is possible to find the coils with the highest discharge activity. Partial discharges can be measured by acoustic methods. The acoustic methods will reveal for example, vibrating coils. An advantage with acoustic methods is that their sensitivity is not dependent on the capacitance of the test object.
Chapter 2
On-line measurements can be made with a rotor mounted scanner. The rotor mounted scanner provides continuous scanning of each slot or tooth in the stator for partial discharges, acoustic emission, vibrations, temperature and the dimension of the air gap between the rotor and the stator. Measuring Vibrations: With a Vibro-meter the air gap between the stator and the rotor as well as stator vibrations can be measured on-line [76]. Temperature: Resistance temperature detectors (RTD) are normally used by embedding them winding of a generator. They will give an indication if the temperature in the winding is higher than normal, causing accelerated thermal ageing [76]. Thermography: An infrared imaging technique to measure the temperature in a generator with a resolution of 1 K. This method is suitable for on-line measurements [82]. There are several diagnostic methods to choose from and they all reveal different kinds of information. It is difficult to perform AC-measurements on a large object due to the large displacement current. In today's insulating materials, mica is an important component because of its endurance to partial discharges. Ironically, this feature of mica insulation makes lifetime prediction more difficult. There is also today a trend towards on-line monitoring for diagnostic purposes either on a periodic or a continuous basis.
10
Electric properties of dielectric materials
Chapter 3
3.1
Introduction
Materials can be divided, with regards to their electric properties, into three main classes, conductors, semiconductors and insulators. In this chapter, the focus will be on insulators (dielectric materials) but a few observations about conductors will be mentioned to make this short review complete. The information presented here should be seen as background knowledge for understanding the dielectric properties observed in measurements and models found in this thesis. This information can be found in a number of different textbooks but the intention was to gather this information and present it in a logical and easy accessible way [12], [18], [49], [64], [66], [74], [75], [88].
3.2
Electrostatics of conductors
If a conductor is put in a static electric field the fundamental property of the conductor implies that the electric field inside the conductor must be zero. Hence, all the charges in the conductor must be located on its surface. The boundary condition at the surface of the conductor states that the electric field must be normal to the surface at every point. It is well known that any electric field applied to a conductor will cause a free flow of charges, which is called current density. This current density, J, will according to Ohms law be proportional to the applied electric field, E. J = E
(A m )
2
(3.1)
where is the conductivity of the conductor. The electron was discovered in 1897 by J. J. Thompson and in 1900 P. Drude proposed a model for electric conductivity of metals based on kinetic gas theory [9]. The basic idea in the Drude model is that there exist freely moving electrons (no interaction) and positive immobile ions built up by the nucleus and its core electrons. The electrons collide with the ions and bounce off in random directions. However, the precise source of scattering does not matter for the qualitative and often quantitative understanding of metallic conduction. If n is the number of free moving electrons per unit volume and the average time between collisions the conductivity can be expressed as
drift
ne 2 = ne m
(1 m )
(3.2)
where e is the charge of one electron and m is the mass of one electron. Instead of talking about average time between collisions, it is common to introduce mobility for the electrons. This is just a different notation but does not change anything in the assumptions made by Drude. 11
Chapter 3
This approach to model conductivity with a drift of charge carriers in an electric field does not only work for metals but is also an important component when modelling semiconductors and dielectric liquids [51]. It could be stated that true conduction in both conductors and in dielectrics arises from free movements of charged particles. Other interesting phenomena which contributes to the current density but are either conduction processes or polarisation processes are electrochemical reactions, diffusion and convection of charge densities. These processes can for example be found in dielectric liquids but also in resins during curing.
3.3
Electrostatics of dielectrics
In the case of a dielectric material, there exist no free charges since positive and negative charges are all bound. The molecules in a dielectric material are usually distributed in such a way that they make the dielectric material overall neutral. When a dielectric material is put in a static electric field, the electric field will penetrate into the material because bond charges cannot move freely. This should be compared to a conductor where an external field cannot penetrate since free moving charges will rearrange themselves so that the internal field will be zero. The internal electric field in a dielectric material will change the material in such a way that the electronic and ionic structure of the molecules on a microscopic scale is shifted. This results in a change of molecular charge density. A multipole expansion of the potential from each molecule with this changed charge density can then be done [9],[49]. It turns out that the dominant molecular multipole with the applied field is the dipole, pn, since most of the molecules are neutral, see Figure 3.1.
10 di pn = qdn -q
+q +q
di+1 -q
Figure 3.1
Molecules are forming dipoles from an applied electric field shifting electronic and ionic charge densities on an atomic scale.
If every dipole location and corresponding charge magnitude were known, a microscopic model could be built. This microscopic model would have electric and magnetic fields, which are very fast varying fields due to the rapid variation in charge distribution on an atomic scale. However, since this is not practical and often not interesting the behaviour of these dipoles is described on a macroscopic scale averaging thousands of molecules and atoms [9],[49]. This is done by introducing the macroscopic electric polarisation vector P which is defined as dipole moment per unit volume. 1 Nv p P = lim i = lim pi Y 0 Y Y 0 Y i=1
(C m )
2
(3.3)
where pi is the dipole moment from the i:th dipole in the unit volume v and N the number of dipoles per unit volume. If the molecules have a net charge of ei and if there is a macroscopic excess of free charges the charge density on a macroscopic level can be written as
12
Electric properties of dielectric materials 1 = lim Y 0 Y

Nv i =1
Chapter 3
+ e
i
excess
(C m )
3
(3.4)
As mentioned above most of the molecules are neutral which results in that the average molecular charge is zero. If the dielectric material is now looked at on a macroscopic level, it is possible to build up the electric field or the potential by linear superposition of the contribution from macroscopically small volume elements V. If we then treat each V as infinitesimal, the potential at x can be written as [18]
1 (x ) = 4
[ (x) - P(x )] dV (V )
x - x
(3.5)
which is actually the potential from a charge distribution (P). This is an electrostatic case which means that E = resulting in that the fundamental postulate for free space must be modified to take into account the electrical polarisation arising in the dielectric materials. The first Maxwell equation is therefore written as
E = 1
0
P]
(C Fm )
2
2
(3.6)
From this the electric displacement is defined as

D=
0
E+P
(C m ) (C m )
2
(3.7)
If the dielectric material is linear and isotropic the electric polarisation is directly proportional to the electric field intensity and can be written as
P=
0 e
(3.8)
where e is the dimensionless electric susceptibility. In a linear dielectric material the electric susceptibility is constant and independent of the magnitude of the electric field intensity and in an isotropic dielectric material the electric susceptibility is constant and independent of the direction of the electric field intensity. The electric displacement can now be expressed as
D=
0
[1 + e ]E =
0 r
(C m )
2
(3.9)
where r is the relative permittivity. If now the dielectric material is also homogenous which means that the electric susceptibility is constant and independent of position in the dielectric material the divergence equation can be written as
E =
0 r
(C Fm )
2
(3.10)
The conclusion of this is that the electric field intensity in a dielectric material (linear, isotropic and homogenous) is reduced by a factor 1/r compared to that in free space [18]. The polarisation of the atoms and molecules in the dielectric material gives rise to electric field intensities inside the material, which partly cancels out the applied electric field intensity. This is used in for example capacitors where the capacitance is increased by r if a dielectric material is inserted between the electrodes. The electrostatic energy in a material is expressed as 13
Chapter 3
We =
1 D E dv = 2V
1 2 V
E 2 dv
(J )
(3.11)
This shows that a dielectric material can store more energy than free space.
3.4
Electric polarisation mechanisms
Electric polarisation in dielectric materials can be produced by many mechanisms in the material. But before looking into these mechanisms it is important to realise the difference between electric polarisation and electric conductivity. It can be stated that [51] "Polarisation arises from a finite displacement of charges in a steady electric field" "Conduction arises from finite average velocity of motion of charges in a steady electric field" The electric polarisation will therefore never contribute to a continuous conduction current unless a very high electric field is applied over the dielectric material. This is good to have in mind when reviewing the following simplified classification of electrical polarisation mechanisms [21], [25], [40], [51], [54], [88] Electronic (optical) polarisation: This occurs when an electric field, EL, in the optical region is applied to an atom with fixed position. The positive nucleus and the negative electron cloud will be shifted relative to each other. The atom requires a dipole moment p, which is proportional to the electric field.
p=
optical
EL
(Cm )
(3.12)
where optical is the electronic polarisability of the atom. Because of the electron shell configuration, the polarisability will most likely be anisotropic but constant in the frequency region studied in this thesis. The electronic polarisability alone is found in nonpolar substances and ranges from around 10-41 Fm2 to 10-39 Fm2. The low values are found for noble gases because they have a completely filled outer electronic shell, which screens the nuclei from any external field. The high values are found for alkali metals with only one electron in the outermost shell. This electronic polarisation alone or interacting with other mechanisms is present in most materials. Molecular (optical and infra-red) polarisation: This occurs when an electric field, EL, is applied to a molecule that has before the field was applied a total dipole moment that is zero. These types of molecules, polar substances, are built up by atoms that are interacting leading to the chemical bonds between atoms and due to symmetry also a distribution of electrons that gives a zero total dipole moment. In this case, the applied electric field will induce a dipole moment due to elastic displacements of charges. Since there is such a great difference between the mass of electrons and nuclei there is one group of normal modes which is related to the displacements of electrons relative the nuclei, for frequencies in the optical range, and one group of normal modes which is related to the displacement of nucleus, for frequencies in the infra-red region. Displacement of a the nucleus also includes displacement of electrons, which is inevitable because of their interaction.
14
Chapter 3
The total dipole moment, p, for polar substances in the frequency range studied in this thesis can be written as
p=(
optical
infra - red
) E
(Cm )
(3.13)
where optical is the electronic polarisability and infra-red is the atomic polarisability of the atom. Both these polarisabilities will most likely be anisotropic but also independent of frequency, electric field and temperature. Examples of polar substances are carbon dioxide CO2, benzene C6H6 and many others both in solid, liquid and gas phase. At sufficient low temperatures many dipolar materials, see below, also start to behave polar like since the thermal energy will be insufficient to turn the dipoles within a reasonable time. The most representative materials of the polar substances are the ionic crystals that may show considerable atomic polarisability. Most crystals of salts like for example sodium Na+Clare examples of ionic crystals. Compared to most other dielectric materials, most salts on melting become ionic conductors. Orientational (optical and infra-red and dipole) polarisation: This occurs when an electric field is applied to a dipolar material. The molecules in the dipolar material, which have a permanent dipole moment, tend to align themselves with the applied field. It is important to emphasise that the electric field has only a small effect on the tendency to alignment and that the thermal effect causing the chaotic rotational motion is dominant. In zero electric field, the molecules will be randomly oriented and the material has no net electric polarisation. The effect of a static electric field on a group of weakly interacting molecules was first studied by Debye [21] see Appendix A. The static polarisability of a dipolar molecule can be written as a sum of orientational, optical and infra-red polarisability
tot 2 p dipole + = 3k T B optical
infra -red
(Fm )
2
(3.14)
where pdipole is the permanent dipole moment of the molecule and T is temperature. Since the assumption made is that the groups of molecules studied are weakly interacting, the local electric field, EL, acting of one molecule will be the same as the applied external electric field, EA. The macroscopic polarisation can therefore be written as
P=
0
N 1)E A = V
tot
EA
(C m )
2
(3.15)
where N is the total number of molecules in the volume V. By changing the temperature, T, and measuring the static relative permittivity, s, it is possible to calculate the permanent dipole moment of the molecule, pdipole, and the sum of its electronic and atomic polarisability, optical+infra-red. So far, interaction has been neglected between molecules, which implies that these models will be inadequate for solid dielectric materials at sufficiently low frequencies where inter-molecular (particle) distances can be very small. A discussion about effects of particle interactions is a topic outside the scope of this thesis. This is a difficult subject but just to mention a historical important step the ClausiusMossotti relation will be mentioned. What they did was to assume that the N isotropically interacting molecules (particles) were confined within a dielectric sphere of volume V. Since the polarisation of a sphere is uniform, it is expected that the mean local electric field, EL, will be the same for all molecules. The local electric field will be the sum of the external electric field, EA, and the average electric field exerted on one molecule from its 15
Chapter 3
(N-1) neighbours. Using the total induced dipole moment for a sphere in a uniform static electric field, the following Clausius-Mossotti relation can be derived
( (
s s
1) E L N = + 2) E A 3V
tot
(3.16)
The vector notation is dropped since EL and EA are parallel. Other expressions relating the local electric field to the external electric field have been derived by Lorentz-Lorenz, Debye, Onsager, Kirkwood and Frhlich and can be found in many standard textbooks [21], [25], [51], [66], [88], [96]. An example of a polar material is water where the atoms (H2O) are arranged in a triangle resulting in a net dipole moment of pdipole=1.84 D (units in Debye: 1 D=3.335610-30 Cm). However, not all molecules are necessarily dipolar, see above, this is very intimately linked with the nature of the chemical bonds between the atoms. Hopping charge carrier polarisation: Another important form of electric polarisation, found mostly in solids for both bulk and surface, is polarisation due to hopping charge carriers. This type of mechanism is something in-between on one hand induced and permanent dipoles and on the other hand free moving charges [51].
Total energy of particle
R12 R21 W pEAcos
x1
x2
Distance
Figure 3.2 A double potential well representing a model for hopping charge carriers. An external electric field, EA, is applied which changes the probability of transition, R21 > R12. Hopping charge carriers (electrons or ions) spend most of their time in localised sites, x1 or x2, see Figure 3.2. They are strongly dependent on temperature (thermal vibrations) and occasionally they make a jump over the potential barrier W. The probability of a hopping transition, R12 or R21, is determined by the distance between two possible locations and the height of the potential barrier W (tunnelling may also occur). If an electric field is applied over the solid dielectric material the probability of transition will change, R21 > R12. It is important to notice that it is not the electric field itself that will cause the charge carrier to jump; it is still the thermal vibrations. If a charge carrier can jump through the whole of the dielectric material then it will contribute to DC current conduction. However, the mobility of the jumping charges will in that case be much lower than free carrier conduction. Hopping charge polarisation is found in electronic and ionic conductors and in and on humid materials like for example glass, mica and granular materials such as sand and soil [51], [53], [54].
16
Chapter 3
Not all dielectric materials involve all the polarisation mechanisms described here. The different mechanisms are also characterised by specific time constants, which may differ by many orders of magnitude. They also depend differently on temperature. There are also other types of dielectric materials for example electrets that have a nonzero net dipole moment and an electric polarisation vector P even when there is no external electric field [12]. Another group of dielectric materials is the ferroelectrics, which also exhibits a net dipole moment without an external electric field. Ferroelectrics show hysteresis when the electric polarisation, P, is plotted versus the electric field, E. Ferroelectricity usually disappears above a certain temperature called the transition temperature [12], [64].
3.5
Dielectric response in time domain
All classical macroscopic electromagnetic phenomena will follow the set of four equations for electric and magnetic fields E and B set up by J. C. Maxwell in 1865 [74], [75].
D = B = 0
(Coulombs law )
E =
B t D H = J + t
(Faradays law ) (Ampres law )
(3.17)
These coupled equations can be seen as the equations giving the electric and magnetic fields everywhere in space provided all the sources, free charge and free current densities and J are given. The constitutive relations, expressing material properties, connect E and B with D and H as
D=
0
E+P
H=
1
0
BM
(3.18)
where P is the polarisation, bond charges, and M is the magnetisation, bond currents, of the material [49]. The dielectric materials studied here are assumed linear, homogenous, isotropic and they have zero magnetisation. If a given electric field E is applied over a dielectric material both the free and bond charges will give rise to sources inside the material in form of charge and current densities which will give rise to a magnetic field B according to Maxwells equations. The current density in the dielectric material will be given by Ampres law as H = E { + E t 1 2 3
0
Induced current
Vacuum displacement current
Polarisation current
P t {
(A m )
2
(3.19)
The first part, the induced current, is the contribution from the materials volume conductivity , the second part is the vacuum displacement current and the third part is the polarisation current. Assume now that the dimensions of the dielectric material are small compared to the change of the electric field E in space. The room coordinates can then be dropped and all field quantities can be seen as only functions of time. The electrical polarisation P can then be divided into two parts, one part representing rapid polarisation processes and one part representing slow polarisation processes [45]. The rapid part follows the applied electric 17
Chapter 3
field whereas the slow part is built up from a convolution integral between the applied electric field and a function called the dielectric response function, f(t). The dielectric response function represents the memory effects in a dielectric material and this function will be discussed further in chapter 3.7. The electric polarisation can now be written as
P (t ) =
0 e E(t ) + P (t ) = 0 e E(t ) + 1 4 24 3 0
Rapid polarisation
0 14 4 4 2444 3 Slow polarisation
f ( ) E(t )d
(C m )
2
(3.20)
The distinction between rapid and slow polarisation processes is not fundamental but depends on the relevant time scale of our observation. Rapid polarisation processes is meant processes which are faster than the time scale of the applied electric field E(t). The total current density J(t) through a dielectric material in an electric field E(t) can then be expressed as.
= t 678 J (t ) = E + 0 (1 + e )E(t ) + f ( )E(t )d { t Induced current 0 444424 144 444443
Total polarisation current
(A m )
2
(3.21)
It is seen from the equation above that the conductivity , the high-frequency component of the relative permittivity and the dielectric response function f(t) will characterise the behaviour of the dielectric material. This gives in the time domain the possibility to apply an electric field, measure the current density and then try to estimate parameters that characterise the material.
3.6
Dielectric response in frequency domain
Assume again that the dielectric material is linear, homogenous, isotropic and non-magnetic. Now only time-harmonic electromagnetic fields, fields that have a time dependence that can be described with one frequency, are considered. The electromagnetic fields can then be written in a complex form where the physical field is represented by the real part. This will result in that time derivatives can be written as the function itself times i [94]. The current density in the frequency domain will then be written according to Ampres law as
H =
Induced current
E {
Vacuum displacement current
i E 02 1 3
Polarisation current
i P {
(A m )
2
(3.22)
If now the same separation of the electric polarisation P in rapid and slow processes is done as in chapter 3.5 the electric polarisation P will be written in the frequency domain as
( P
)=
( E
( )= ) + P
( E
)+
( f
( ) )E
(C m )
2
(3.23)
The convolution integral describing slow polarisation processes in the time domain will be in the frequency domain represented by a product. This important simplification will make calculations easier and faster. Now the dimensionless frequency-dependent complex electric susceptibility () is introduced and is defined as [51] 18
Chapter 3
( ) = f(t)e -i t dt ) = ( ) i ( ) = f
0
(3.24)
The real and the imaginary part of the complex electric susceptibility () can be seen as the cosine and the sine transform of the dielectric response function f(t).
( (
) = f (t ) cos( W ) dt ) = f (t ) sin( W ) dt
0 0
(3.25)
Carrying out the inverse cosine and sine transforms will return f(t). f(t) can be determined from either the real or imaginary part of the electric susceptibility.
f (t ) = f (t ) =
2 2
( ) cos( W ) d
0
( ) sin( W ) d
0
(1 s )
(3.26)
This is interesting because this shows that the Fourier transform is a link between the frequency-dependent electric susceptibility () and the time-dependent dielectric response function f(t). Given one of the functions then the other can be determined. The total current density J() in a dielectric material under a time-harmonic excitation E() can be expressed as
( J
)=
( E
)+
( i E
)+ i
( E
( )) = ) + 0 ( ( ) i ( )) E
= Conduction loss } Dielectric loss 6 7 8 6 7 8 ( = i 0 1 + e + ( ) i + ( ) E 1 4 4 2 4 4 3 Capacitive part 0 1444 4 2444 4 3 Resistive part
(A m )
2
(3.27)
From this expression it is seen that there is one part of the current density J() which is in phase and one part which is 90 degrees before the driving time-harmonic electric field E(). The part of the current that is in phase with the driving field is associated with the energy losses in the dielectric material. Two types of energy losses are seen in the material. The first type, which is due to the conduction (free charges) in the material, gives rise to ohmic losses. The second type, which is due to electric polarisation in the material, gives rise to what is called dielectric losses. Dielectric losses occur due to the inertia of the bound charges when they are accelerated in the driving field. The part of the current that is 90 degrees before the driving field, displacement current, is associated with the capacitance of the material. In many situations, it is more convenient to talk about the complex relative permittivity, which is defined as follows [51]
19
Chapter 3

6 7 8 ( ) = 1 + e + (
( J
)= i
( ) ( ( ) i ( )) E
)=
0
) ( )
(3.28)
It is seen from the equation above that the conductivity , the high-frequency component of the relative permittivity and the electric susceptibility () characterise the behaviour of the dielectric material under time-harmonic excitation. This equation shows, as in the time domain, that it is possible in the frequency domain to make measurements that characterise the material. Under the assumptions that the dielectric material is linear, homogenous and isotropic, the measured information in either the time domain or frequency domain are equal. The information found in one domain can be transformed via the Fourier transform of f(t) or () to the other.
3.7
Time-dependence of dielectrics
The inevitable inertia of all physical processes in nature is one of the most obvious reasons for a time-dependent dielectric response. No dielectric material is able to directly follow arbitrarily changing forces (driving electric field). This can be compared to the instantaneous response of free space. To go a step further in the modelling of electric polarisation in time-dependent fields it is necessary to better define the dielectric response function f(t) [51]. If an electric field with the amplitude is applied over a dielectric material for t seconds the time-dependent electric polarisation can be expressed as
P(t) =
0
E t f(t)
(C m )
2
(3.29)
where f(t) is the dielectric response function. Causality demands that there should be no reaction before action therefore
f(t) 0 t < 0
(1 s )
(3.30)
If the assumption is made that the dielectric material considered has no permanent or persistent polarisation the following is required
lim f(t) = 0
t
(3.31)
Assume now that the dielectric material is linear. Then the electric polarisation from consecutive delta functions is the sum of electric polarisation from each individual delta function. This is called the principal of superposition. The easiest way to understand this is to study the total electric polarisation in a dielectric material from an applied electric field consisting of a few delta functions separated in time and with different amplitudes like in Figure 3.3. Then, according to Figure 3.3, the total electric polarisation in the dielectric material at time t can be written as
P(t) =
0
t (E1f(t t 1 ) + E 2 f(t t 2 ) + E 3f(t t 3 ) )
(C m )
2
(3.32)
where f(t) is the dielectric response function of the dielectric material. To extend this further in an arbitrary electric field the polarisation can be written as a sum of delta functions of different amplitudes. 20
Chapter 3
E t
E t
Polarisation
Dielectric response function, f(t), of the material
Time
Figure 3.3
An excitation of a dielectric material from three different delta functions of height E1,2,3 (V/m) and duration t. The total polarisation at time t is according to the principal of superposition the sum of polarisation from each delta function.
When the number of delta functions goes to infinity, the total electric polarisation can be expressed as a convolution integral between the dielectric response function and the applied electric field.
P(t) =
f ( )E(t )d
0
(C m )
2
(3.33)
The lower limit of the convolution integral is set to zero since f(t)=0 for t<0. The physical interpretation of a convolution integral is that the dielectric material has a memory of its history. This means that the electric polarisation at time t=t1 depends on the history of the electrical field from t= up until t=t1. The memory effect in a dielectric material, which is stored in f(t), can vary from a few fractions of a second up until days depending on the material. From the expression of the electric polarisation the polarisation after an infinite time under a static electric field E0 can be expressed as
P(t = ) =
E 0 f ( )d
0
(C m )
2
(3.34)
From this expression it is obvious that the dielectric response function f(t) must be finite since the steady state polarisation is finite. Each dielectric material has its own dielectric response function that can change in different ways depending in which time window the material is studied. Typical response functions representing slow polarisation processes that are found in electrical insulation materials can be seen in Figure 3.4. The dielectric response function of the classical Debye model can be written as [21]
f (t ) = e t
(1 s )
(3.35)
The Debye expression represents a physical process that is hardly seen in solid dielectric materials. It was derived under the assumption that dipoles are not interacting which is true 21
Chapter 3
only for a number of polar liquids but not for solid materials. Attempts to produce solid Debye-like material by diluting the dipolar part to remove interactions do not give result since the magnitude of loss most often decreases more rapidly than the strength of interaction [51].
log(Dielectric response function)
Curie - von Schweidler
"General response" Slope -m Slope -n
Debye
t=
log(Time)
Figure 3.4
Different types of dielectric response functions, f(t), in time domain. For polar liquids response functions like the Debye function are commonly found. In solid dielectrics, the response functions are more of the fractional power law type as seen in the General response and Curie von Schweidler type of functions.
The dielectric response function of the Curie-von Schweidler model is valid for many dielectric materials over a wide range of times [51]. f (t ) = A t n
(1 s )
(3.36)
It is important to notice that this model diverges at zero time if n>1. This means that the model with n>1 is not applicable for arbitrarily short times. If n<1 then the integral representing the total charge stored in the material will also diverge this is physically impossible. This means that the model is not valid for long times if n<1. In spite of these two limitations experiments show that the Curie-von Schweidler model, in specific time ranges with different values of n, fit measured data in most of the cases very well. The general response model represents a transition at t= between two different processes in the dielectric material. Depending on the value of n the material can be classified as a charge carrier system, 0<n<1, or a dipolar type of system, 1<n<2 [51].
f (t ) =
(t )
A + (t
)n
(1 s)
(3.37)
Another also useful expression with nice analytic features that behaves asymptotically in the same way as the general response is
m t f (t ) = A e t + (1 e t
t )
0 < m <1 0<n<2
(1 s )
(3.38)
As with the Curie-von Schweidler model this model diverges for short times with m>1 and at 22
Chapter 3
long times for n<1. This dielectric response function is of course derived from observations in measured data and is not derived from any physical model. However, it is a useful model for engineering purposes and it is applicable over a wide range of times [38]. The dielectric response function can also be described in frequency domain via a Fourier transform of the time domain expression as seen in chapter 3.6. Physical processes can be interpreted from both time and frequency domain since they are equivalent. However, it is sometimes more popular to interpret data in frequency domain mostly because a few physical processes appear as peaks, compare Figure 3.4 and Figure 3.5. In Figure 3.5 (A), the frequency response of the Debye model is shown. This frequency response can be calculated analytically from equation (3.35) as
(
)=
1+ i
(3.39)
In Figure 3.5 (B), a typical dipolar loss peak found in solid materials is shown. This peak is wider than the Debye peak and has a low frequency behaviour n and a high frequency behaviour -m. The symmetric case, n=m<1, can be found in weakly doped ionic crystals and the asymmetric case, n>m, in polymers and glasses as - and -relaxation peaks and also in some polymers at cryogenic temperatures [51]. Debye log() log() " -2 log() Figure 3.5 log() -1 Dipolar " n Symmetric: n = m < 1 Asymmetric: n > m -m
(A) Frequency response of a Debye model typically found in gases and dilute liquids with non-interacting dipoles. (B) Wider dipolar peak found in solid materials like polymers.
The wider dipolar peak corresponds in time domain to the dielectric response functions found in equations (3.37) and (3.38). The response function in (3.37) can not be Fourier transformed but the response function in equation (3.38), which behaves asymptotically the same, can be Fourier transformed to [90]
(
) = A
(1
n (1 n ) (1 m ) 1 m (1 + i )1n +i )
(1 n ) (i )1n
0< m <1 0< n<2
(3.40)
In Figure 3.6 (A), the frequency response from hopping charges is shown. This type of frequency dependence is mostly seen in charge carrier dominated systems, which will show strong dispersion in both bulk and interfacial regions and will for the lower frequencies show low frequency dispersion (LFD) where -n, n<1 [54]. In Figure 3.6 (B), the flat frequency response or frequency independent response is shown which is the response when no other mechanism is dominant [51]. The flat response can be derived from the Fourier transform of 23
Chapter 3
the time domain Curie-von Schweidler model found in equation (3.36). The Fourier transform of equation (3.36) is
( - n) ) = A 1-n (i ) 0< n <1
(3.41)
and then when n1 a flat frequency response will occur. Hopping charges (LFD) -n log() log() n1 " log() Figure 3.6 log() " Flat response n-1 cot(n /2)
(A) Frequency response from hopping charges found in many charge carrier dominated systems. (B) Flat response found in all materials when no other mechanism dominates.
In measurements there are often combinations of these mechanisms described above found both along the frequency axis but also when changing the temperature.
3.8
Kramers-Kronig relations
If the complex electric susceptibility is an analytic function in the upper half of the -plane, it is possible to relate the real and imaginary part of the electric susceptibility to each other on the real axis via Cauchys theorem [19]. These relations, Hilbert transforms, are called the Kramers-Kronig relations and can be expressed as [8], [51]
) = 1 lim a
1
+a
(x) dx x -a
+a
(x) ( ) = lim dx a x a
(3.42)
Since the complex electric susceptibility is obtained via a Fourier transform of the real dielectric response function f(t), see equation (3.24), the following symmetry relation must be valid for the electric susceptibility (
) = ( )
(3.43)
This symmetry relation leads to that the Kramers-Kronig relations can be rewritten in the following way
24

( ) = 2 lim
a 0 a
Chapter 3
x (x) dx x2 2
a
( ) =
(x) lim 2 dx a x 2 0
(3.44)
This form of the Kramers-Kronig relations are the ones with the most physical significance since it is only possible to measure data at zero and positive frequencies. The Kramers-Kronig relations are a direct consequence of the assumption that the dielectric material studied here can be described by a system that obeys causality. A general meaning of causality is that there is no reaction before there is an action in the system, see equation (3.30). An other interesting relation which follows directly from the Kramers-Kronig relations is that the real part of the electric susceptibility for static conditions, =0, can be written as
( = 0) =
lim
+a
-a
+a (x) 2 dx = lim (x) d(lnx) a x 0
(3.45)
which relates the total area under the imaginary part of the complex electric susceptibility plotted against natural logarithm of the frequency to the value of the real part of the complex electric susceptibility at zero frequency. This shows that to every polarisation mechanism there must exist a corresponding dielectric loss peak somewhere in the frequency spectrum. The real part of the complex electric susceptibility is almost independent of frequency when the dielectric loss is small, see Figure 3.7.
(), "() (0) 1() r 1 1"() 2() "()
2
r 2
p1
p2
log()
Figure 3.7
Two different polarisation mechanisms, p1 and p2, with corresponding loss peaks. The real part of the electric susceptibility at zero frequency represents in a log-log plot the total area under the imaginary part of the electric susceptibility. This leads to the real part of the electric susceptibility being almost constant in-between two loss peaks.
The Kramers-Kronig relations are very useful for checking measurement data of the complex relative permittivity in the frequency domain. Measurement data from any test object should obey the Kramers-Kronig relations otherwise, there is reason to suspect that something is wrong with the measurement or the measurement set-up [30], [51].
25
Chapter 3
When making frequency domain measurements the complex relative permittivity is measured. Recalling equation (3.28), the complex relative permittivity can be expressed as
)=
+ (
) i
+ ( )
(3.46)
where () is the complex electric susceptibility. The high-frequency component of the relative permittivity, , is according to Figure 3.7 chosen to r1 or r2 depending on which polarisation mechanism is measured. Applying the Kramers-Kronig relations to this measured set of data it is seen that a constant value, like , and a 1/ behaviour, like /(0), will not contribute to the result of the transform. This can be shown as follows [90]
) = 1 lim a
+a
+a
-a
dx =
lim [ln x
a
+a -a
a lim ln a a +
=0
(3.47)
for a constant and

(
)=
lim
a
-a
x(x
dx =
0
x lim ln a x
= -a
+a
a lim ln a a +
=0
(3.48)
for a 1/ behaviour. These results have the consequence that in order to get the true KramersKronig compatible pair from measured data the high-frequency component of the relative permittivity, , must be subtracted from the real part of the complex relative permittivity and the pure DC conductivity, /0, must be subtracted from the imaginary part of the complex relative permittivity. What the Kramers-Kronig relations are really saying is that there is a relationship between the real and the imaginary part of the complex electric susceptibility. The complex relative permittivity is the Fourier transform of the dielectric response function. For a few dielectric response functions, it is possible to express the Kramers-Kronig relations analytically. One example is the Curie-von Schweidler model, see Appendix B, where the Kramers-Kronig relations can be expressed as [106]
f(t) = A t -n (
- n) ) = A 1-n (i )
( (
) = cot n ) 2
0 < n <1
(3.49)
where n-1 is the slope of both the real and imaginary part of the complex electric susceptibility in a log-log plot. When 0<n<0.5 the imaginary part of the complex electric susceptibility is higher than the real part and when 0.5<n<1 the opposite is true. A special case is when n=0.5 where the real and imaginary parts are equal. For most of the solid dielectric materials used as insulators, low loss materials, n1. Another example is the Debye model, which can be expressed as [21] f (t ) = e t ( ( (
)=
1+ i
)= )
(3.50)
However, it is important to note that the Debye model is for most of the cases not valid in solid dielectric materials. In spite of this the Debye model can be interesting from a mathematical point of view since it is sometimes possible to expand measured data, both in the time and frequency domains, as a sum of different Debye models. This sum has an
26
Chapter 3
analytic expression for both the Fourier transform and the Kramers-Kronig relations. This type of curve fitting is an easy way of avoiding calculating time consuming transforms [30].
3.9
Heterogeneous dielectrics
Most insulation systems found in practical applications are composites or mixtures of several different dielectric materials. The calculation of the dielectric properties of such a medium is a problem of both theoretical and practical importance. The principal aim is to calculate the relative permittivity of the mixture in terms of the relative permittivities of the constituents, their relative amounts and their spatial distribution. It is important to realise that each dielectric material has a dielectric response and when putting these materials together the total response will not only reflect each material but also the way they are put together. Sometimes the total dielectric response describes one physical process but the materials in the mixture have dielectric responses revealing totally different physical processes. This can be the case when for example a high viscosity liquid with ionic conduction is mixed with a solid, which has no conduction [2]. Two simple systems are found in Figure 3.8 representing a capacitor filled with two different dielectrics in two different ways representing a parallel and series case [41], [88], [96].
(A) S1 h S2 (B) S Cp1 Cp2 h1 h2 Cs1 Cs2
1 2
Figure 3.8
A capacitor which is filled with two dielectrics, 1 and 2, but distributed in two different ways (A) in parallel and (B) in series.
In the first case, Figure 3.8 (A), there are two dielectrics where each form a cylinder with a cross section, of an arbitrary but uniform cross section, with its axis parallel to the applied electric field. This case of parallel dielectrics can be generalised and the effective relative permittivity for N different dielectrics in parallel can be written as
eff
= ( i wi )
i =1
(3.51)
where wi is the volume ratios of the ith dielectric material in the capacitor in (A). In the second case, Figure 3.8 (B), there are two dielectrics that instead are in parallel and in the same manner as above can this case be generalised. For N different dielectric materials in series the effective relative permittivity is expressed as
1
eff
N wi = i =1 i
(3.52)
where also here wi is the volume ratios of the ith dielectric material in the capacitor in (B).
27
Chapter 3
However, in many practical cases composite dielectrics are complex (statistical) mixtures of several dielectrics, which are both in parallel and in series. In this case the equivalent diagrams in Figure 3.8 are not sufficient but the true value of the effective relative permittivity should lie somewhere between the values determined by equations (3.51) and (3.52). This is formulated in the Wiener inequality [88], [96]
1 N wi i =1 i
eff
( i wi )
i =1
(3.53)
There exist many semi-empirical formulas for the calculation of the effective relative permittivity of statistic mixtures [81]. A few will be mentioned here just to illustrate their complexity. Lichtnecker-Rother have derived the formula for logarithmic mixing
log(
eff
) = (w i log( i ))
i =1
(3.54)
which works well for foams and porous material. Landau-Lifshitz derived for statistic mixtures the following formula [66]
13 13 = (w i ( i ) ) eff ) N i =1
(3.55)
and Maxwell-Wagner derived the well known formula for a binary mixture consisting of dielectric spheres, relative permittivity 1, distributed uniformly in a dielectric of relative permittivity 2. The effective relative permittivity can then be written as
eff
( 1) 1 2w 2 ( 2 2 + 1) = 2 ( 2 1) 1 + w (2 + ) 2 1
(3.56)
where w is the volume concentration of the dielectric spheres. From this expression, the wellknown Maxwell-Wagner effect (dispersion) can be derived by making the spheres conductive and the surrounding dielectric media insulating. In Figure 3.9 the different ways of modelling the effective relative permittivity for two dielectric materials, as mentioned above, are plotted as a function of their volume ratios in the mixture. This illustrates in a good way the Wiener inequality where the extreme cases are given by the series and the parallel models. The models presented above are semi-empirical formulas that work well especially for statistic mixtures. In this thesis two different resin-rich mica tapes have been studied which are used as electrical insulation in rotating machines. These tapes consist of epoxy, mica and a carrier material which is polyester film (PET) or woven glass. Both tapes have a given internal periodic structure, which is of a dimension not vanishing small compared with the final insulation dimensions. Therefore it is possible to identify a small unit cell which can be described by a network of series and parallel capacitors which contains the dielectric materials the tape is built up from [39].
28
Chapter 3
eff Landau-Lifshitz
Lichtnecker-Rother Serie
Parallel
Maxwell-W agner
0% A 100% B
100% A 0% B
Figure 3.9
Effective relative permittivity for several different ways of mixing two dielectric materials A and B with relative permittivity A and B as a function of their volume content in the mixture.
From a more general point of view it turns out that a H-structure like the one found in Figure 3.10 is interesting not only for resin-rich mica tapes but also for other insulation combinations found in high voltage apparatus. In chapter 7, there will be a more in-depth discussion and the effective relative permittivity for the network shown in Figure 3.10 will be solved both in time and frequency domain.
1
S1 h1 hG h1
S1
C1 Cp2
2 3
C2
1 2
2 1
C2
C1
Figure 3.10
Example of an H-network formation used to describe the dielectric behaviour of resin-rich mica tape insulation.
3.10
Temperature dependence of dielectrics
In general, the dielectric losses caused by dipole mechanisms reach a maximum at certain definite temperatures Tm [96]. The increase in temperature and resulting decrease in viscosity will affect the dielectric losses originating form rotational movements. On one hand the degree of dipole orientation will increase with temperature, total polarisation will increase, which will increase the dielectric losses but on the other hand there is a reduction in energy to overcome the resistance of the viscous medium (internal friction) for the rotating dipoles decreasing the dielectric losses with temperature. For a quite large group of solid dielectric materials, it is found that the shape of the dielectric response (spectral shape) does not change very drastically with temperature. This is at least 29
Chapter 3
true for temperature ranges over which the material does not change its internal structure too much (phase transitions etc.) [51]. This means that it is sometimes possible to normalise the measurement data for different temperatures by shifting the time/frequency spectra so that they coincide into one single curve called the master curve [30], [51]. The resulting master curve covers a larger time/frequency range than one single temperature run, thus containing more information and also adding increased reliability in the measurement points. The master curve technique can be used both in the time and frequency domains. In the time domain the dielectric response function can be shifted, without changing shape, in both amplitude and time using the following expression
f (t, T2 ) = A y (T1 , T2 ) f (A x (T1 , T2 ) t, T1 )
(3.57)
where Ax(T1,T2) and Ay(T1,T2) are the shift in time and amplitude when going from temperature T1 to T2. This expression can then be Fourier transformed to get the corresponding shift in the frequency domain
A y (T1 , T2 ) ( T2 ) = , T 1 A (T , T ) A (T , T ) x 1 2 x 1 2
(3.58)
The function Ax,y(T1,T2) is most often strongly temperature dependent and can for some dielectric materials be expressed with the Arrhenius factor [26]
A x,y (T1 , T2 ) = e
E x,y 1 1 kB T2 T1
(3.59)
where T1 and T2 represent starting and ending temperatures, Ex,y is the activation energy of the material and kB is Boltzmanns constant. For materials that are more complex, it is also possible that the activation energy is temperature dependent. To illustrate the master curve technique for an Arrhenius activated dielectric material both in the time and frequency domains the dielectric response has been plotted for three different temperatures see Figure 3.11.
log(Dielectric response function)
3 2 1
log(Complex susceptibility)
f(t,T=T ) f(t,T=T ) f(t,T=T )
(A)
( ,T ) ( ,T )
1 2
(B)
( ,T )
3
"( ,T )
1
T >T >T
3 2
T >T >T
3 2
"( ,T ) "( ,T )
2 3
log(Time)
log( )
Figure 3.11
Possible temperature dependence for an Arrhenius activated dielectric material in both the time domain (A) and frequency domain (B). Here the material has Ax(T1,T2)=Ay(T1,T2) according to equations (3.57) and (3.58).
30
Chapter 3
In this case Ax(T1,T2)=Ay(T1,T2) according to equations (3.57) and (3.58), giving only a shift along the frequency axis in the frequency domain which is the case for most Arrhenius activated materials. In the frequency domain, master curve shifting can be done with both the real and imaginary part of the complex electric susceptibility since they theoretically contain the same amount of information. If in the frequency domain the complex relative permittivity (or complex capacitance) is measured it is important to first subtract the contribution in the real part from the infinity part, , and in the imaginary part the conduction, /0, see equation (3.28). The subtraction of these two parts should be done for each temperature run in such a way that the remaining real and imaginary parts are Kramers-Kronig compatible. The risk is that if the high-frequency component of the relative permittivity, , and conductivity, , are not subtracted, their corresponding temperature dependence will corrupt the master curve for the complex electric susceptibility. The master curve technique is not only an useful tool for analysing the temperature dependence for dielectric materials but can also be used for studying pressure, humidity and other processes which will affect the rate processes in the system without changing the shape of the dielectric response spectra.
31
Dielectric measurement techniques
Chapter 4
4.1
Introduction
Diagnostic measurements of high voltage insulation in for example high voltage cables, transformers or rotating machines can be done either in the time or frequency domains. Today existing commercial diagnostic methods are in time domain measurements of polarisation/depolarisation currents and recovery voltage as a function of time and in frequency domain measurements of capacitance and loss, tan, as a function of frequency. Parameters that are used to describe a linear, homogenous and isotropic insulation material are in the time domain the high-frequency component of the relative permittivity, , conductivity, , and dielectric response function, f(t), and in the frequency domain complex relative permittivity (), see Figure 4.1. Time domain Pol. & depol. currents
Recovery voltage
, , f (t ) (
Dielectric material
Capacitance & loss Frequency domain Figure 4.1 Dielectric materials can be diagnosed in either in time or frequency domains. It is important to understand how the different methods in the two domains work and how they relate to each other especially under realistic conditions.
To make reliable diagnostics it is important to understand how well each diagnostic method can characterise a high voltage insulation system depending on the insulations dielectric response. Further, it is also important to understand the mathematical relations between each diagnostic method and how measured data can be transformed from one method to the other. In this chapter, three different electrical insulation systems will be studied at room temperature with the three different methods mentioned above. The insulation systems are oil/paper insulation, resin-rich mica tape insulation and cross-linked polyethylene (XLPE) insulation. These three insulation systems are commonly used today within high voltage applications and they all have different dielectric responses ranging from higher, oil/paper, towards lower loss, XLPE. It will be shown what type of information is realistic to extract not only from each method but also from each insulation system. Conclusions will be made about 33
Chapter 4
the best choice of method depending on insulation system response and on the type of information and physical mechanism one wants to diagnose.
4.2
Polarisation and depolarisation currents
To measure polarisation and depolarisation currents is one way in the time domain to investigate the slow polarisation processes, defined in chapter 3.5, in a dielectric material [51]. This means measuring the rate of charges which are transported to the electrodes of a test object when a voltage U(t) is applied over the same electrodes. The current through a test object with the geometric capacitance C0, can according to equation (3.21) be expressed as
C i(t ) = 0 8 (t ) + 0
0
dU (t ) + dt
t d f ( )U (t )d dt 0
(A )
(4.1)
where is the high-frequency component of the relative permittivity, is the conductivity and f(t) is the dielectric response function of the dielectric material. Assume now that the test object is totally discharged and that a step-voltage is applied with the following characteristics
t<0 0 U (t ) = U 0 0 t t 1 t > t1 0
(V )
(4.2)
This will give a zero current for times before t=0 and a so called polarisation current for times 0tt1. The polarisation current due to the step-voltage U0 is built up by two parts. One part is related to the conductivity of the test object and one part is related to the activation of the different polarisation processes within the test object. In dielectric materials used as high voltage insulation there are many different polarisation processes which all have different time constants widely distributed in time. The polarisation current is expressed as
i pol (t ) =
+
0
(t ) + f (t ) C0 U 0 > 0
for 0 t t 1
(A )
(4.3)
where (t) is a delta function originating from the derivative of the applied step voltage at t=0. The current for times t1<t is called the depolarisation current. This depolarisation current is built up by the relaxation of the polarisation processes that were activated by the step-voltage U0. The depolarisation current is expressed as
i dpol (t ) = (f (t ) f (t t 1 )
(t t1 )) C 0 U 0 < 0
for t 1 < t <
(A )
(4.4)
where -(t-t1) is the delta function originating from the negative voltage step at t=t1. Shifting the time, t=t-t1, to the same time base as the polarisation current the depolarisation current can be written as
i dpol (t ) = (f (t ) f (t + t 1 ) +
(t )) C0 U 0 < 0
for 0 < t <
(A )
(4.5)
The measured depolarisation current has a negative sign compared to the polarisation current but when they are plotted together, the sign is changed on the depolarisation current.
34
Chapter 4
Measuring polarisation and depolarisation currents was in this thesis done with a setup consisting of three ROSS high voltage relays, switching time approximately 0.5 seconds, a Keithley 242 3kV/6mA DC source and a Keithley 6517 electrometer [35], [37]. The measurement system will measure the polarisation and depolarisation currents according to equations (4.3) and (4.4) with the exception of the contribution from the delta function. To measure the delta function and to find the high-frequency component of the relative permittivity , it is better to measure charge instead of current at least in the initial stage of the measurement. Limiting factors for the quality of the measurement data in this setup is the switching time, rise time of the DC source and the fact that this type of measurement is a wideband measurement which makes it sensitive to noise and complicates the use of filtering algorithms [38]. All measurements were done at room temperature, T=23C. In Figure 4.2 the measured polarisation and depolarisation currents at U0=3 kV for a piece of a new oil/paper insulated HVDC cable, geometric capacitance C0=144 pF (total capacitance C=640 pF), are plotted against time for charging and discharging periods of t1=105 seconds. The currents have been scaled with U0C0 so that the result is comparable with other insulation systems of different geometry and charging voltage. A guard configuration was arranged at both ends of the cable consisting of two metal cylinders that were pressed into the insulation at the ends in-between the high voltage conductor and outer lead screen (measurement electrode). The ends were then covered with polyethylene film and sealed with tape to reduce oil leakage and moisture penetration.
Pol/Depol currents scaled by U C [1/s]
10
-1
C = 144 pF
0
Slope n = -0.37
10-2
U0 = 3 kV Slope n = -0.31
10-3
I
10-4
pol
Idpol Slope n = -1.05
10
-5
10-1
100
101
102
103
104
105
Time [s]
Figure 4.2
Polarisation and depolarisation currents in a new HVDC cable with oil/paper insulation, scaled with geometrical capacitance, C0, and charging voltage, U0. A clear contribution from conduction current is seen after around 6000 seconds in the polarisation current.
The polarisation and depolarisation current response for oil/paper insulation seen in Figure 4.2 is a large response in magnitude for being an insulating material. The slopes of the currents in log-log scale are moderate indicating rather slow time dependence. However, there are two break points the first around 10 seconds and the second around 2000 seconds where the slopes change indicating a change in the polarisation processes. The influence of external noise is low but a tendency to noise in the polarisation current is seen at the end of the charging period, which comes from the ripple, U, in the charging voltage. Around 6000 seconds there is a split up between the charging and discharging currents indicating that the conduction current starts to be dominant over the polarisation current. 35
Chapter 4
In Figure 4.3 the measured polarisation and depolarisation currents, at U0=3 kV, are plotted for a new factory produced stator bar insulated with resin-rich mica tape made for a hydropower generator [16]. The stator bar has a geometric capacitance C0=130 pF and a total capacitance C=620 pF and was charged and discharged for a period of t1=105 seconds. The currents have also here been scaled by U0C0. A 2-3 mm guard with an overlapping equipotential surface pushing down the potential lines in the gap was applied at both ends of the measuring electrode (semi conducting tape) together with field grading pads [44], [104], [105]. It is difficult to make a good guard with good electrical contact since the stator bar has rectangular geometry. The stator bar was then put inside a grounded metal tube to reduce the influence from external noise.
Pol/Depol currents scaled by U C [1/s] 0 0
10-1
C = 130 pF
0
10
-2
U0 = 3 kV
10-3 10-4 10
-5
Slope n = -0.91 I Idpol

pol
10-6 10
-7
10-1
100
101
102
103
104
105
Time [s]
Figure 4.3
Polarisation and depolarisation currents in a new stator bar, a part of the winding in a hydropower generator, scaled with geometrical capacitance, C0, and charging voltage, U0.
It is directly seen that the response from the stator bar is smaller in magnitude and has a different shape than the response from the oil/paper insulated HVDC cable above. The stator bar has only one slope, n=-0.91, in log-log scale showing a faster time dependence than the oil/paper insulation. A weak tendency to a separation between polarisation and depolarisation currents around 2104 seconds indicates a possible conduction current. At the end of the polarisation current measurement, there is also ripple-induced noise from the voltage source as for the HVDC cable. In Figure 4.4 the measured polarisation and depolarisation currents, at U0=3 kV, are plotted for a new XLPE insulated 12 kV power cable. The XLPE cable has a geometric capacitance C0=814 pF and a total capacitance C=1.87 nF and was charged and discharged for a period of t1=105 seconds. The currents have also here been scaled with U0C0. At the ends a guard was applied in the same manner as for the stator bar together with a high voltage stress cone (geometrical) [44], [104], [105]. The ends were then put into grounded metal boxes to screen the high voltage and thereby reduce the coupling to the high voltage side. From the three test object studied here the XLPE cable has the lowest magnitude of response and the steepest slope, n=-1.19 in log-log scale. The steep slope, fast time dependence of the response, of the XLPE cable indicates that most of the polarisation processes take place at higher frequencies. An offset current of about 10.210-13 A was measured in the XLPE cable for zero applied voltage, U0=0 V. The offset was found in the XLPE cable even though it was 36
Chapter 4
short circuited for 7 days at room temperature prior to the measurement. This offset current was subtracted from the measured values before plotting in Figure 4.4. The influence of the ripple in the charging voltage is clearly seen in the XLPE polarisation current. From the polarisation current with its noise and the depolarisation current with its possible influence of an offset current it is difficult to identify any conduction current at all.
Pol/Depol currents scaled by U C [1/s] 0 0
10-3 10
-4
C = 814 pF
0
10-5 10 10
-6
Slope n = -1.19 I
U0 = 3 kV
pol
-7
Offset level I
10-8 10-9 10
-10
dpol
10-1
100
101
102
103
104
105
Time [s]
Figure 4.4
Polarisation and depolarisation currents in a new XLPE insulated cable scaled with geometrical capacitance, C0, and charging voltage, U0. The currents show fast time dependence and a strong influence of external noise at the end of both charging and discharging periods. A measured offset current 10.210-13 A was subtracted from both the polarisation and the depolarisation currents.
4.2.1
Estimation of the dielectric response function
In order to estimate the dielectric response function f(t), see equation (3.29), from a depolarisation current measurement it is assumed that the dielectric response function is a continuously decreasing function in time. If this is true (not always true) then the test object can be charged for a sufficiently long time, t1, so that
f (t ) >> f (t + t1 ) for t > 0
(4.6)
Recalling equation (4.5), the following approximation of the dielectric response function can be made, disregarding the contribution from the delta function f (t ) = i dpol (t ) C0 U 0 + f (t + t 1 ) i dpol (t ) C0 U 0 for t > 0
(1 s)
(4.7)
where idpol(t) is the measured depolarisation current shifted to start at t=0. This is probably the easiest way to determine the dielectric response function f(t), but it is also very time consuming from a measurement point of view [30], [35], [51]. Many solid dielectric materials have dielectric response functions, which decrease slowly with time resulting in that the assumption above is time consuming to fulfil. There is an approximate rule of thumb which says that the test object should be charged for at least ten times as long as the measurement time of the depolarisation current in order to get a depolarisation current which is proportional to the dielectric response function [51]. 37
Chapter 4
In Figure 4.5 the dielectric response function is estimated from depolarisation current measurements for the oil/paper HVDC cable, the resin-rich mica tape stator bar and the XLPE AC power cable, however disregarding the finite charging period t1=105 seconds.
Dielectric response function, f(t) [1/s]
10-2
HVDC cable, oil/paper Statorbar, resin-rich mica
10
-4
10-6
10-8
AC power cable, XLPE
10
-10
10-1
100
101
102
103
104
105
Time [s]
Figure 4.5
The dielectric response functions, f(t), calculated from measured depolarisation currents for the HVDC cable, stator bar and AC power cable without compensating for the finite charging period, t1=105 seconds.
It is also possible to compensate for the finite charging period, t1, by trying to estimate the contribution of f(t+t1) in equation (4.7). One possible way to do this is by approximating f(t) for times t>t1 with a t-n behaviour that was curve fit to the last decade of measured data. In Figure 4.6, the results from compensating for the finite charging period are shown for the HVDC cable and the stator bar. No compensation was made for the XLPE cable since its dielectric response function for long times was corrupted with a lot of noise and an offset current, see Figure 4.4, making the compensation sensitive.
Dielectric response function, f(t) [1/s]
10-1 10-2 10
-3
HVDC cable, oil/paper Statorbar, resin-rich mica B
10-4 10
-5
A A B
10-6 10
-7
10-1
100
101
102
103
104
105
Time [s]
Figure 4.6
The dielectric response functions for the HVDC cable and the stator bar, with, curve A, and without, curve B, compensation for the finite charging period. Compensating measurement data saves valuable measurement time.
The compensation made, affects both the HVDC cable and the stator bar but only the measured data in the last decade of the charging period, which is in good agreement with the 38
Chapter 4
rule of thumb mentioned above. This way of compensating the measured data is a good way of saving measurement time.
4.2.2
Estimation of the conductivity
From the measurements of the polarisation and depolarisation currents, it is possible to estimate the DC conductivity, , of the test object. If the test object is charged for t=t1 equations (4.3) and (4.5) can be combined to express the DC conductivity as
=
0
C0 U 0
(i (t ) + i (t )) f (t + t )
pol dpol 0 1
for t > 0
(1 m )
(4.8)
It can sometimes be very difficult in solid dielectric materials to distinguish between true DC conduction and different types of polarisation mechanisms. It is important to realise that the test object must be charged until the effects from the dielectric response function have disappeared, f(t+t1)<</0, in order to get the true DC conductivity. A common way to estimate the DC conductivity is to use any of the following three expressions. The first expression, 1, uses the polarisation current only and neglects the influence of the dielectric response function f(t) see equation (4.3).
1
C0 U 0
i pol (t )
for t > 0
(1 m )
(4.9)
The second expression, 2, use both the polarisation and depolarisation currents but neglect the influence of the shifted dielectric response function, f(t+t1) see equation (4.8).
2
C0 U 0
(i (t ) + i (t ))
pol dpol
for t > 0
(1 m )
(4.10)
The third expression, 3, uses the polarisation current and subtracts the dielectric response function, fcomp(t) calculated from the depolarisation current and compensated for the finite charging period.
3
C0 U 0
(i (t ) f (t ))
pol comp
for t > 0
(1 m )
(4.11)
In Figure 4.7, the results from estimating the DC conductivity for the HVDC cable and the stator bar at room temperature are shown using the three ways described in equations (4.9) to (4.11). No estimation is done for the XLPE cable because of noise and the offset current see Figure 4.4. New XLPE is such a low loss material that it is difficult to talk about conductivity at room temperature. For the HVDC cable the conductivity is found to be around 510-16 1/m but for the stator bar it is more difficult and a weak assumption for the conductivity could be 410-18 1/m. An important conclusion is that it is more time efficient and accurate to estimate the conductivity by using both polarisation and depolarisation currents.
39
Chapter 4
10-14 10-15

HVDC cable, oil/paper Conductivity, [1/m]
10-16
Conductivity, [1/m ]
Statorbar, resin-rich mica
10
-17
10
-15
10-18
10
-16
= 510
-16
S/m
= 410-18 S/m
103 104 105 10
-19
103
104
105
Time [s]
Time [s]
Figure 4.7
Estimated conductivity for the HVDC cable and the stator bar using the polarisation current, 1, both polarisation and depolarisation currents, 2, and using polarisation and depolarisation currents compensating for the finite charging period, 3.
4.3
Recovery voltage
The recovery voltage method is another method in the time domain to investigate slow polarisation processes, which were defined in chapter 3.5 [14], [15], [20], [45], [54], [80], [97]. A step voltage, U0, is applied over the electrodes of a completely discharged test object with geometric capacitance C0, see Figure 4.8.
Charging U
0
Grounding
Open circuit
Recovery voltage, U (t)

R
Time
Depolarisation current, i (t)

d
Polarisation current, i (t)

p
Figure 4.8
During a recovery voltage measurement the test object is charged with U0 from 0 t t1, grounded from t1 t t2 and for t>t2 the recovery voltage is measured during open circuit conditions.
During the charging period there will be a current ipol(t) through the test object. After the charging period, the test object will be short-circuited for a period. During this grounding period there will be a current idpol(t) through the test object. When the grounding period is finished, the recovery voltage UR(t) is measured under open circuit conditions. The source of 40
Chapter 4
the recovery voltage is the relaxation processes inside the dielectric material, giving rise to an induced charge on the electrodes of the test object. The easiest is to look at the current through the test object, see equation (4.1) to understand why there is a recovery voltage building up during open circuit conditions. The currents for charging, grounding and open circuit can be written as
i pol (t ) = +
0
(t ) + f (t ) C0 U 0
0 t t1 t1 < t < t 2 t2 < t <
i dpol (t ) = (f (t ) f (t t 1 ) i R (t ) = 0
(t t1 )) C0 U 0
(4.12)
where f(t) is the dielectric response function of the dielectric material. The charging period will activate the different polarisation processes. Depending on how long the test object is charged, t1, different polarisation processes with different time constants will be activated. Then during the grounding period, the different polarisation processes will start to relax. Depending on how long the test object is grounded, t2, a different number of polarisation processes are totally relaxed. Totally relaxed means that there is no information left about the polarisation process. Then during the open circuit period the polarisation processes, which were not totally relaxed during the grounding period, will relax and give rise to a recovery voltage over the electrodes of the test object. The recovery voltage will re-polarise the dielectric material as it is building up to a degree that is in relation to the magnitude of the recovery voltage. This re-polarisation makes the interpretation of the recovery voltage somewhat difficult since what is done is that the relaxation of polarisation processes in the dielectric material (test object) is integrated over the dispersive material itself. The recovery voltage can be expressed with the following equations
i R (t ) = 8 R (t ) +
t
dU R (t ) + dt
U 0 (f (t ) f (t t 1 )) +
(4.13)
U R (t = t 2 ) = 0
d f (t )U R ( ) d = 0 for t 2 < t < dt t 2
where UR(t) is the recovery voltage. The forward problem is to calculate the recovery voltage UR(t) for a given excitation knowing the conductivity , high-frequency component of the relative permittivity and dielectric response function f(t). This is a straightforward problem and can be analytically solved for a dielectric response function that follows the Debye model [29], [35], [37]. The inverse problem is to calculate , and f(t) from a given recovery voltage knowing the excitation. This is much more difficult problem than the forward problem.
4.3.1
The forward problem
If the conductivity , the high-frequency component of the relative permittivity and the dielectric response function f(t) are known for the test object it is possible to at least numerically calculate the recovery voltage UR(t) from any given excitation. It is possible for a 41
Chapter 4
few cases depending on the properties of the dielectric response function to even express the recovery voltage analytically.
4.3.1.1
Equations giving the analytic expression for the recovery voltage
Assume now that the test object is charged with a step-voltage U0 for 0<t<t1, followed by a grounding period t1<t<t2. The recovery voltage can then be expressed according to equations (4.12) and (4.13). The expression for the zero current, iR(t)=0, in equation (4.13) can be rewritten with the help of Leibnitzs rule for differentiation of integrals as [90]
U R (t ) +
0
dU R (t ) + U 0 (f (t ) f (t t 1 )) + f (0)U R (t ) + dt
d f (t )U R ( )d = 0 for t 2 < t < dt t2
(4.14)
Make now the assumption that the dielectric response function f(t) has the property that its derivative can be written as
d f (t dt
) = A(t ) B( )
(4.15)
Put this assumption into the equation (4.14) and take the derivative with respect to time. This will result in a second order differential equation describing the recovery voltage. =7 b(4 t ) 4444 =7 a (4 t ) 448 64444 4 8 6444 2 d U R (t ) f (0 ) A (t ) dU R (t ) f (0 ) A (t ) f (0 ) + + + + 2 U R (t ) = A(t ) 0 dt A(t ) dt 0 d A(t ) = 0U0 dt (f (t ) f (t t 1 )) A(t ) (f (t ) f (t t 1 )) 144 444444424444444443 = t t2 < t < (4.16)
U R (t = t 2 ) = 0
U (f (t 2 ) f (t 2 t 1 )) dU R (t = t 2 ) = 0 dt
This differential equation is difficult to solve because it has coefficients that are time dependent. Nevertheless, this time dependence disappears if a (t ) = a A(t ) = constant A(t ) b(t ) = b (4.17)
The above equations show that there exists an analytic solution for the recovery voltage described by the differential equation in equation (4.16) if the dielectric response function fulfills the conditions in equations (4.15) and (4.17). These two conditions are strong and will reduce the number of possible dielectric response functions that might fulfill them. The differential equation above can be solved with a standard method like Method of Inverse Operators and the solution can be written as [72], [89], [90]
42

U R (t ) = m1 = e -m1t

Chapter 4

1 a a 2 4b 2 1 m 2 = a + a 2 4b 2
( (
m 2 m1 + C1e -m1t + C 2 e -m2 t
t e m1t dt e -m2 t
t e m2 t dt
t2 < t <
) )
(4.18)
with a, b and (t) from equation (4.16) and constants C1 and C2 are given by the initial conditions in the same equation. One dielectric response function that will fulfil the conditions in equations (4.15) and (4.17) is the response function given by the Debye model [21], [25], [51]. The recovery voltage from an arbitrary step-voltage excitation of a Debye model will be described according to equation (4.18) by the sum of two exponentials with different time constants. Experimental results show that it is common that the recovery voltage can be described by a sum of exponentials with different time constants although the dielectric response function cannot be described by a Debye model.
4.3.1.2
Numerical calculation of the recovery voltage
For most of the dielectric response functions f(t) it is not possible to analytically calculate the recovery voltage UR(t) from an arbitrary excitation of the test object. When this is the case, a numerical routine must be used to calculate the recovery voltage [35], [37]. Assume again that the test object is charged with a step-voltage U0 for 0<t<t1, followed by a grounding period t1<t<t2. Integrate now equation (4.14) between ti and ti+1 where ti+1> ti>t2 giving
t i +1
U R (t )dt +
ti
t i +1 0
dU R dt + dt ti
t i +1 0U0
(f (t ) f (t t ))dt +
1 ti
t i +1
d t dt = 0 t i+1 > t i > t 2 f ( t ) U ( ) d R dt ti t2
(4.19)
which can be rearranged as

t i +1
0 ti t i +1
(t)dt +
(U R (t i +1 ) U R (t i ) ) + U 0 (f(t) f(t t 1 ) )dt + ti ti
t i +1
(4.20)
f(t
ti
i +1
8 R ( )d + (f(t i+1 f(t i )U R ( )d = 0

t2
t i+1 > t i > t 2
If the integrals between ti and ti+1 are approximated with the trapezoidal formula the recovery voltage UR(ti+1) can now be expressed as the following recurrence formula
2 t t i i +1 2U 0 f(t 1 ) U R (t i ) t t i +1 i 2 + + f(t 0 ) t i +1 t i 0
t i +1
U R (t i +1 )
(f(t) f(t t ) )dt

1 ti
43
Chapter 4 2 t i+1 t i
t2
(f(t
2
ti
i +1
f(t i )U R ( )d t i , t i+1 > t 2 +

0
(4.21)
t i+1 t i
+ f(t 0 )
This recurrence formula contains two integrals one of which depending on the dielectric response function could be solved analytically and one that must be approximated with some approximate formula for definite integrals [90]. The chosen step size hi=ti+1-ti will at step i generate an error that will be added to all the other errors from the previous steps. Experimental calculations show that the step size should be chosen to h<10 seconds in order to reduce the influence from the errors at each step. The result is also very much dependent on how the approximation of the dielectric response function is made at t=t0 and how well the grounding period t2-t1 is determined from the measurement. Calculating the recovery voltage and comparing this with a measured recovery voltage is a good test of the given dielectric response function f(t), conductivity and high-frequency component of the relative permittivity .
4.3.1.3
Recovery voltage measurements
Measuring recovery voltage was in this thesis done with the same setup as for measuring polarisation and depolarisation currents but with the difference, that after the grounding period the voltage was measured during open circuit conditions [35], [37]. The measurement system measured the polarisation current during the charging period but not the depolarisation current during grounding because of limitations in the electrometer switching between current to voltage mode [61]. Limiting factors for the quality of the measurement data in this setup is the relay switching time, rise time of the DC source, input impedance of the voltage measuring device and the importance of a well working guard configuration on the test object [35], [37]. All recovery voltage measurements presented in this thesis were made with a driven guard coupling [60], [61]. The measurements were done at room temperature, T=23C. In Figure 4.9, three different recovery voltage measurements are shown for the oil/paper insulated HVDC cable. The cable was always charged with 500 V for a charging period of tc=100, 500, 1000 seconds and grounded for tg=10, 10, 10 seconds respectively. The solid lines are the numerically calculated recovery voltages (forward problem) from solving equation (4.13) using the estimated dielectric response function and conductivity (510-16 1/m, 4.1) from chapters 4.2.1 and 4.2.2. The agreement between measured and calculated recovery voltage is good, the only critical parameters in the calculations are the accuracy of the measured grounding time, t2-t1, and the estimation of the dielectric response function for short times, f(t0). The estimation of the dielectric response function becomes more crucial the steeper the slope of the function is on a log-log scale. For the oil/paper system, the recovery voltages can reach large magnitudes also in comparison with the charging voltage, which will result in that the re-polarisation of the material during the recovery voltage period becomes significant. A peak value is reached, slightly shifted to longer times for longer charging times, after which the conduction current becomes visible and starts to reduce the recovery voltage.
44

100
Chapter 4
t / t =1000/10
c g
Calculated data Recovery voltage [V]

80
Parameters : C 0 = 144 pF
t / t =500/10
c g
60
U 0 = 500 V = 5.0 10 -16 1 = 4.1 P
40
t / t =100/10
c g
f (t ) = see Figure 4.5
20
0 0 10
101
102
103
104
Time [s]
Figure 4.9
Measured compared to calculated recovery voltage for the oil/paper insulated HVDC cable. Charging was done with 500V for tc=100, 500, 1000 seconds and grounding for tg=10, 10, 10 seconds respectively. Note the large response and the clear influence of conductivity starting to reduce the recovery voltage at around 2000 seconds.
In Figure 4.10, three recovery voltage measurements are shown for the resin-rich mica tape insulated stator bar. The stator bar was charged with 500 V for a charging period of tc=100, 500, 1000 seconds and grounded for tg=7, 7, 7 seconds respectively. There are also here good agreement between calculated recovery voltages, solid lines, and the measured ones. The stator bar has lower recovery voltage response than the oil/paper insulated HVDC cable, which was also the case for the measured polarisation and depolarisation currents. No clearly seen peaks are observed in this measurement interval, which indicates that there is no influence of conduction current. Note in equation (4.13) that the magnitude of conduction current is proportional to the magnitude of recovery voltage.
20 Calculated data Recovery voltage [V] 15 t / t =500/7
c g
t / t =1000/7
c g
Parameters : C 0 = 130 pF U 0 = 500 V = 3.0 10 -17 1 = 4.3 P
10 tc/ tg=100/7 5
0 0 10
101
102 Time [s]
103
104
Figure 4.10
Measured compared to calculated recovery voltage for the resin-rich mica tape insulated stator bar. Charging was done with 500V for tc=100, 500, 1000 seconds and grounding for tg=7, 7, 7 seconds respectively. There are no obvious signs of any influence of conduction current.
45
Chapter 4
In Figure 4.11, three recovery voltage measurements are shown for the XLPE insulated AC power cable. The XLPE cable was charged with 3 kV for a charging period of tc=100, 500, 1000 seconds and grounded for tg=1, 1, 1 seconds respectively. The charging voltage was raised to 3 kV and the grounding time was shortened to 1 second because of the otherwise very small recovery voltage response.
0.5 Calculated data Recovery voltage [V] 0.4 0.3 0.2 0.1 0 100 tc/ tg=100/1 tc/ tg=500/1 t / t =1000/1
c g
Parameters : C 0 = 814 pF U 0 = 3000 V = 3.0 10 -19 1 = 2.3 P
101 Time [s]
102
103
Figure 4.11
Measured compared to calculated recovery voltage for the XLPE insulated AC power cable. Charging was done with 3 kV for tc=100, 500, 1000 seconds and grounding for tg=1, 1, 1 seconds respectively. The peak and following drop in the measured recovery voltages could wrongly be interpreted as conduction current but is in reality an offset current in the test object.
The measured and calculated recovery voltages in Figure 4.11 do not agree so well. This suggests that influences like offset currents in the test object, leakage currents over the guard and a low input impedance in the electrometer are factors affecting the quality of the measured recovery voltage data. In Figure 4.12 it is illustrated how an offset current in the range of 0.1 pA, see Figure 4.4, will affect the recovery voltage measurement.
0.4 Calc. U , =3.0 10
R -19
Parameters :
S/m Calc. UR, =10
-14
C 0 = 814 pF U 0 = 3000 V
t / t =100/1
c g
Recovery voltage [V]
S/m
0.3
0.2
= 2.3
0.1
Meas. U
U Offset =
2 3
Meas. UR + Calc. UOffset
i Offset t C XLPE cable
(V ) with (A )
0 10-1
10
10 Time [s]
i Offset 1 2 10 13
10
10
Figure 4.12
The effect from an offset current on the measured recovery voltage, UR(t), for the XLPE cable. The offset current was chosen to be in the range of 0.1 pA according to what was measured in Figure 4.4. 46
Chapter 4
The measured recovery voltage can be wrongly interpreted as conductivity, see calculation made with a conductivity of 10-14 S/m. The conclusion is that it is difficult to measure recovery voltage on a low loss material like XLPE.
4.3.2
The inverse problem
The inverse problem, calculating the dielectric response function, conductivity and highfrequency component of the relative permittivity from a known excitation and the measured recovery voltage is difficult. Attempts have been made to assume a certain analytic form of the dielectric response function and then using a non-linear constrained minimisation routine to minimise the current in equation (4.13). This will give a set of parameters as a result including conductivity and high-frequency component of the relative permittivity [36], [37]. A general problem with all minimisation routines is that they are very much dependent on the initial starting values to avoid ending up in a local minima. However, this method works well when the shape of the dielectric response function is known and can be described with a simple parameterised analytic function. An example of this will be given below in chapter 4.3.2.1. Another approach is to expand both the dielectric response function and the recovery voltage in a set of basis functions chosen in such a way that the convolution integral in equation (4.13) can be solved analytically giving a linear system of equations [37]. The problem with this approach is that many dielectric response functions are very rapidly changing functions over many decades (large dynamics) giving with this approach a linear system of equations with a coefficient matrix that is badly conditioned. It would be easier to solve the inverse problem if the charging and discharging currents together with the recovery voltage were measured but then the motivation to measure the recovery voltage and not the easier interpreted polarisation/depolarisation currents would be hard to make.
4.3.2.1
Constrained minimisation of the current
In this case, oil/paper insulation was studied and an analytic dielectric response function was chosen that was known to fit measured data. The general response function was chosen and parameterised as f (t; p 0 , p1 , p 2 , p 3 ) =
(t
p 1 ) + (t p1 )
p2
p0
p3
(1 s )
(4.22)
with p 0 , p1 > 0 , p 3 > p 2 > 0 and p 3 > 1 The strategy is to find the parameters pn so that the current iR(t) in equation (4.13) is equal to zero for all t>t2. To find the parameters pn a non-linear constrained minimisation routine based on a Sequential Quadratic Programming method written by The Math Works Inc. (MATLAB 4.2a) was used to minimise the square of the current iR(ti) for a number of ti>t2 i=0,1,...,m [73]. The problem to be solve can be stated as min[ j(t, p ) j(t, p )] subjected to constrains G(p ) 0
p
(4.23)
where G(p)0 are the constraints chosen for the parameters pn. The constrains are related to physical properties of the system like causality and that no permanent polarisation exists 47
Chapter 4
which was briefly discussed in chapter 3.7. The times ti at which the current iR(t) is calculated are chosen to be linearly spaced throughout the recovery voltage measurement. It is important to find a good starting approximation. If the starting approximation is poor, the calculations will take many iterations to converge and can be sensitive to numerical problems such as truncation and round-off errors. In order to find a good starting approximation equation (4.13) was simplified by omitting the convolution term. This simpler problem then generated a reasonable initial approximation. There will also be faster convergence if an analytic gradient of the function being minimised is supplied to the minimisation routine. The function, which is minimised, and the constraints must be continuous functions in order to obtain a solution. Another limitation is that the solution generated might be a local minimum. To be sure that a global minimum has been found it is recommended to minimise the problem several times using different initial starting points. In Figure 4.13, the calculated dielectric response functions by minimising the current in equation (4.13) are shown from two different recovery voltage measurements made on an oil/paper insulated HVDC cable. It was chosen to use two recovery voltage measurements, one with a fairly short charging period of 100 s and grounding period of 10 s and one with a longer charging period of 1000 s and grounding of 500 s. In the same figure is also shown a calculated dielectric response function that is achieved by minimising the square of the currents for both the short and the long recovery voltage measurements. The "short" recovery voltage measurement (tc/tg=100/10) gives more information about the fast polarisation processes than the long recovery voltage measurement (tc/tg=1000/500) that contains more information about the slow polarisation processes. When we combine the two recovery voltage measurements, we will get a dielectric response function, which will contain information about both the slow and the fast polarisation processes.
Dielectric response function, f(t) [1/s] 10
-2
10
-3
Measurements:
fR1+R2(t) f (t)
U R1 (t ) : t C t G = 100 10
R2
U R2 (t ) : t C t G = 1000 500
10
-4
Meas. f(t) Calc. from U (t)

R2 R1 R1
Calc. from U (t) Calc. from U (t) + U (t)

R2
f (t)
R1
10
-5
10
-1
10
10
10 Time [s]
10
10
10
Figure 4.13
Calculated dielectric response function from minimisation of the current from two different recovery voltage measurements made on a oil/paper insulated HVDC cable.
In Figure 4.14, the self-consistency is checked of the minimisation procedure by calculating back the recovery voltages from the found dielectric response functions. It is seen that when calculating back UR(t) using the same tc/tg as was used in the measurement, from which the dielectric response function was calculated, the fit is good. However, when calculating the 48
Chapter 4
other recovery voltage it is seen that tc/tg=1000/500 has a better fit to tc/tg=100/10 than the other way around. This is due to that the tc/tg=100/10 is charged for a relatively short period of time, which will result in that it contains very little information about the slow polarisation processes. On the other hand it contains more information about the fast polarisation processes which will result in that it will fit the tc/tg=1000/500 in the beginning as seen in the figure.
120
Calc. U (t) from f (t)
R1 R2 R2 R1 R1 R1 R2 R2
100 Recovery Voltage [V] 80 60
Calc. U (t) from f (t) Calc. U (t) from f (t) Calc. U (t) from f (t)
Parameters : C 0 = 144 pF U 0 = 1000 V = 2.7 10 -16 1 P
Measured data
40 20 0 1 10
2 3 4
f R1 (t ) = see Figure 4.13
= 4.1
f R2 (t ) = see Figure 4.13

10 10 Time [s] 10
Figure 4.14
Calculated recovery voltage from two different dielectric response functions compared to measured. The first response function was calculated (minimised) from the short recovery voltage measurement and the second response function from the long.
In Figure 4.15, it is seen that when the dielectric response function that is determined from the two, short and long, recovery voltage measurements is used the fit to the measured data is much better.
120 100 Recovery Voltage [V] 80
Measured data Calc. UR1(t) Calc. UR2(t)
Parameters : C 0 = 144 pF U 0 = 1000 V = 2.7 10 -16 1 P f R1+ R2 (t ) = see Figure 4.13
60 40 20 0 10
1
= 4.1
10
10 Time [s]
10
Figure 4.15
Calculated recovery voltage from one dielectric response function compared to measured. The response function was calculated (minimised) from two, short and long, recovery voltage measurements.
In the calculations made so far the DC conductivity is assumed to be known. In Figure 4.16, dielectric response functions have been calculated using different values of the DC 49
Chapter 4
conductivity. What is seen is that when conductivity is increased the slope of the dielectric response function for longer times will be steeper than the measured slope. When, in comparison, the conductivity is decreased, the slope of the dielectric response function will be more flat than the measured slope. The calculated response functions in Figure 4.16 are clearly different but are all satisfying equation (4.13) but with different DC conductivities.
Dielectric Response Function [1/s] 10
-2
Parameters : C 0 = 144 pF
10
-3
U 0 = 1000 V
= 4.1
10-4
Meas. f(t)
= normal = 4.5*normal = 0.45*normal 10-5 0 10 10

1
U R2 (t ) : t C t G = 1000 500
10
10 Time [s]
10
10
Figure 4.16
Calculated dielectric response functions from a recovery voltage measurement with different values of the conductivity, .
By calculating back the recovery voltage curve used to find the dielectric response function and comparing with the measured data see Figure 4.17 it is seen that the highest conductivity is clearly too high to be consistent with the measurement. This gives an upper limit for the conductivity and shows that it is also possible to make a simple estimation of the conductivity in the test object from a recovery voltage measurement.
120 Meas. UR2(t) 100 Recovery Voltage [V] 80 60 40 20 0 1 10 102 Time [s] 103 104 Calc. = normal Calc. = 4.5*normal Calc. = 0.45*normal
Parameters : C 0 = 144 pF U 0 = 1000 V f R1+ R2 (t ) = see Figure 4.5
= 4.1
Figure 4.17
Calculated recovery voltages from using different dielectric response functions that were calculated with different conductivity.
To measure the polarisation current during the charging period of a recovery voltage measurement will add additional information in the determination of the dielectric response function and the DC conductivity of the test object [37].
50
Chapter 4
4.3.3
Memory effects
It is important to understand the influence on the test object from a recovery voltage measurement. The memory effects from one recovery voltage measurement can be expressed as the polarisation of the test object at time t>t3 when starting the new measurement. This polarisation can be written as P(t) = t1 E 0 0 f (t 0
)d
+ f (t
t2
t3
) ER ( ) d
E0
t > t3
(C m )
2
(4.24)
where E0 is the static field caused by the step-voltage U0 applied to the test object from time 0<t<t1 and ER(t) is the recovery electric field during the measurement from t2<t<t3. The test object is grounded after t=t3 waiting for the next measurement. From equation (4.24) it can be seen that the recovery voltage period will contribute to the polarisation (memory) in the test object for times t>t3. When comparing this to the polarisation in the test object from a polarisation and depolarisation current measurement where ER(t)=0, it can be seen that the memory effects from a recovery voltage measurement are stronger. During the measurement period, the recovery voltage will polarise the test object. Since the recovery voltage, depending on the properties of the dielectric material in the test object can reach considerable voltage levels it is not certain that this polarisation from the recovery voltage period can be neglected. In order to investigate the memory effects from a recovery voltage measurement three consecutive measurements were performed. The test object was a guarded piece of new resinrich mica tape insulated stator bar, insulation thickness 5 mm and C0=31 pF. It was before the first measurement completely discharged at 90 C for 24 hours. The three recovery voltage measurements were then performed with different short circuit times, TS, before each measurement see Figure 4.18 (A). This was then compared to a similar set of polarisation and depolarisation current measurements see Figure 4.18 (B).
120 100 Recovery voltage [V] 80 60 40 20 0 10-1 U0 = 200 V, T = 90 C t = 500 s
1 2
(A) 10 Pol/depolarisation currents [A]
(B)
-9
U = 200 V, T = 90 C
0
t = 2500 s
c
t = 2500 s
g
#1, Ipol TS>2500 s
t =1s
#1, U #2, U
R R
10
-10
#1, I
dpol
T >2500 s
S
T > 6000 s
S S
#2, Ipol, TS= 2500 s #2, Idpol, TS= 2500 s #3, Ipol, TS= 1 s
T = 6000 s
#3, UR TS= 1 s
100
101
102
103
104
10
-11
#3, Idpol, TS= 1 s
10-1
100
101
102
103
104
Time [s]
Time [s]
Figure 4.18
Memory effects from previous measurements of (A) recovery voltage and of (B) polarisation and depolarisation currents. Between measurement #1 and #2 the test object was short circuited for the total measurement time of #1 compared to between #2 and #3 where the test object was short circuited for 1 second (TS=short circuit time before measurement). 51
Chapter 4
There was a significant difference between measurement #3 and measurements #1 and #2 for the recovery voltage method, something that was not observed for the polarisation and depolarisation currents method. Experimental results show that when measuring recovery voltage it is important to discharge all memory effects before starting a measurement. A simple rule of thumb is that the test object should be grounded for at least as long as the previous measurement (charging period + recovery voltage period) before starting the new measurement.
4.4
Loss factor and capacitance
An equivalent method in the frequency domain to study the slow polarisation part as defined in chapter 3.5 is to study the response from a sinusoidal excitation [28], [37], [51], [77], [98], [104]. Since only sinusoidal excitations are considered, the Fourier transform is applicable to equation (4.1) resulting in the following expression
IM (
) = C0 (
0
( U A
)+ i
( U A
)+ i
( ( )U A
))
(4.25)
where C0 is the geometric capacitance, () the electric susceptibility, the conductivity and the high-frequency component of the relative permittivity of the test object. This expression can be rewritten in the following way
IM (
)= i
C0
+ (
) i
+ ( ) U A (
)= i
C 0 ( (
A( ) ) i ( ))U
(4.26)
where () is the complex relative permittivity. Assume now that a sinusoidal voltage is applied to a test object and the total current through the test object is measured. This could be illustrated like in Figure 4.19 below where the voltage vector has been chosen to have zero phase angle. Im IM IC

UA IR
t
Re Figure 4.19 A voltage UA is applied over a lossy dielectric material and the current IM is measured. IM is built up from a capacitive part, IC, and a resistive part, IR, from which the dimensionless quantity tan is calculated.
Then the complex relative permittivity at the frequency of the applied field, assuming a capacitive test object, can be calculated as
52
Chapter 4
) i ( ) =
I ( 1 M ( i C0 U A
)= )
i IM i e & 0 UA
IM (cos( & 0UA
) i sin ( ))
(4.27)
It is important here to notice that the imaginary part of the complex relative permittivity (loss part) contains both the resistive losses (conduction) and the dielectric losses (polarisation). When making a measurement at an arbitrary frequency it is not possible to separate these two losses. However, the resistive part is often dominant for very low frequencies. The behaviour of the complex relative permittivity when the resistive losses are dominant is that the imaginary part of the complex relative permittivity has a slope of -1 and the real part of the complex relative permittivity is constant, see equation (4.26). It is then possible, from the measured imaginary part of the complex relative permittivity, to determine the conductivity of the test object. Another way of presenting the measured information is to study the frequency dependent ratio between the imaginary and real part of the complex relative permittivity called tan [51].
( (
) = sin ( ) = I R ) cos( ) I C
= tan
(4.28)
One practical advantage with tan is that it is independent of the geometry. This makes tan the only possible parameter to study when the geometry of the test sample is not known. However, when geometry is known, the real and imaginary part of the complex relative permittivity contain separately more information (conductivity, high-frequency component of the relative permittivity and electric susceptibility) than the ratio between them. It is also sometimes argued that it is easier to interpret data in the frequency domain, capacitance and tan as function of frequency or temperature, than measured data in the time domain since frequency data sometimes show nice features like peaks [26], [30], [51]. The complex relative permittivity was in this thesis measured with a high voltage setup based on the principal of balancing out the capacitive part of the current before measuring it [28], [37], [98], [104], [105]. With this setup, voltages up to 20kV/20mA in the frequency range 0.1 mHz to 1 kHz could be used. Limiting factors for the quality of the measurement data in this setup are the high voltage divider, balancing capacitor, phase shift of anti-aliasing filters (matched channels for voltage and current) and the performance of the electrometer [37], [105]. All measurements were done at room temperature, T=23C. In Figure 4.20, the real and imaginary part of the complex electric susceptibility measured at 3 kV peak are plotted for the new oil/paper insulated HVDC cable. It was chosen to plot the data so that the real and imaginary parts would fulfil the Kramers-Kronig relations see equation (3.44). This was done because it is sometimes easier to identify different polarisation mechanisms doing so [51], [54]. Looking at the oil/paper insulated cable it can be seen that it has rather high losses, like in the time domain. The insulation also showed a strong frequency dependence and there are no visible problems with external noise but it is difficult to distinguish a DC conduction current since the cable shows a LFD (low frequency dispersion) behaviour for low frequencies identified by the real and imaginary parts being parallel [51], [54]. Interesting features for oil/paper insulation is the position of the minima in frequency and the slope of the curves in the LFD region, which can be related to for example the temperature of the material [30].
53
Chapter 4
10 Complex electric susceptibility
1

C0 = 144 pF U 10
0 peak
= 3 kV
Slope n = -0.69
10
-1
" "
10
-2
10
-4
10
-3
10
-2
10
-1
10
10
10
10
Frequency [Hz]
Figure 4.20
The complex electric susceptibility is plotted for the oil/paper insulated HVDC cable. There is no clear contribution from a conduction current for the lower frequencies in this frequency window (because and are parallel).
In Figure 4.21, the real and imaginary part of the complex electric susceptibility measured at 200 V peak are plotted for the new resin-rich mica tape insulated stator bar. It was also here chosen to plot the data so that the real and imaginary parts would fulfil the Kramers-Kronig relations. The stator bar has lower losses than the oil/paper insulated cable for frequencies lower than 0.1 Hz, which was also the case in the time domain. It also has very moderate slopes for both the real and imaginary parts no visible influence of external noise and there is no indication of DC conduction current (indicated by a -1 slope). The frequency response of the stator bar also shows a weak possible dipolar peak centred on 0.1 Hz.
100 Complex electric susceptibility C = 130 pF
0
Upeak = 0.2 kV
10
-1
Slope n = -0.06
" "
10
-2
10
-4
10
-3
10
-2
10
-1
10
10
10
10
Frequency [Hz]
Figure 4.21
The complex electric susceptibility is plotted for the resin-rich mica tape insulated stator bar. The stator bar has in frequency domain moderate slopes for both and .
In Figure 4.22, the real and imaginary part of the complex electric susceptibility measured at 3 kV peak are plotted for the new XLPE insulated AC power cable. The real and imaginary parts are plotted so that they fulfil the Kramers-Kronig relations. The XLPE cable has very low losses and weak frequency dependence (low dispersion). Due to noise and a very small 54
Chapter 4
signal, it was not possible to measure lower frequencies than 1 mHz. There are no visible signs, like a -1 slope for the losses in frequency, indicating DC conduction current.
10 Complex electric susceptibility
-1
C = 814 pF
-
10
-2
peak
= 3 kV
10
-3
Slope n = 0.17
" "
10
-4
10-4
10-3
10-2
10-1
100
101
102
103
Frequency [Hz]
Figure 4.22
The complex electric susceptibility is plotted for the new XLPE insulated AC power cable. The cable has very low losses and weak frequency dependence. It was not possible due to noise and signal amplitude to measure for lower frequencies than 1 mHz.
A general remark concerning the above frequency domain measurements of complex relative permittivity is that it is possible to reach information about higher frequencies than in the time domain (TD starts from around 1 Hz). This is of course dependent on the two measurement setups used here but when it comes to high voltage measurements this is often the case.
4.5
How to choose the best measurement method
The dielectric diagnostic methods studied in this thesis, polarisation/depolarisation currents and recovery voltage in the time domain and complex relative permittivity (capacitance and loss) in the frequency domain are all methods commonly used today. It is important to know what kind of information each method gives and what limitations each method respectively has. Here a few conclusions will be drawn regarding choice of measurement method from the measurements made on the oil/paper HVDC cable, resin-rich mica tape stator bar and the XLPE cable. These three insulation materials represent not only three different levels of dielectric losses but also three different dielectric responses. In the time domain, today, it is often difficult to reach higher frequencies than around 1 Hz due to the bandwidth of the current measuring electrometer, mechanical switching of high voltage and rise time of the high voltage source. Since time domain measurements are wideband measurements, they are sensitive to noise making it difficult to measure on low loss materials like XLPE. Time domain measurements work best for high loss systems and systems with significant DC conduction currents. Frequency domain is very convenient for both low and high loss systems when it comes to measuring at higher frequencies. Since frequency domain measurements are a narrow-band measurement, filtering gives good noise performance especially for higher frequencies. If the insulation material is linear in voltage, it is an advantage to lower the applied voltage to reach 55
Chapter 4
higher frequencies. When it comes to determining DC conductivity features the frequency domain measurements can be time consuming since they require in most of the cases data from at least 2 cycles. In Figure 4.23, a schematic drawing has been made to illustrate the area of interest for the two domains. The loss-frequency behaviours from the three systems studied here have been added to the figure to relate a system to a possible method. The XLPE system is such a low loss that it is preferable measured in the frequency domain. The resin-rich mica tape system and the oil/paper system are preferable measured in the frequency domain for the higher frequencies. Time domain "DC" Oil/paper Resin-rich mica tape XLPE "Low Loss" Frequency domain
"High Loss"
Frequency Figure 4.23 Schematic drawing of how to chose measurement domain, time or frequency, depending on the magnitude of losses in the dielectric material and frequency range of interest.
Time domain works well for the oil/paper system but also for the resin-rich mica tape system, especially at temperatures higher than room temperature, when determining slow polarisation or DC conductivity. The difficult "corner" to measure is on a low loss material for low frequencies with high voltage. The basic idea is to choose a diagnostic method and domain according to the degree of dielectric loss in the system but also according to the time or frequency range a certain interesting phenomena is expected to be found in. Time and frequency domains represent two simple possibilities of excitation voltage. There are perhaps different combinations in the excitation voltage, which will provide possibilities to extract more information for a given time than described by these methods presented here.
4.6
Relations between time and frequency domain
It is important, for a linear high voltage insulation material, to be able to transform measured data from one domain to the other. It is especially common to transform data from time to frequency domain since it is in many cases easier and more accepted to interpret data in frequency domain than in time domain [26], [51]. A schematic drawing is shown in Figure 4.24 describing the theoretical relations between the measurement methods discussed in this thesis. It is assumed here that the high voltage 56
Chapter 4
insulation materials are linear and the applied driving voltage, in time domain a step voltage and in frequency domain a sinusoidal voltage, have been applied for a sufficiently long time. Time domain Pol. & depol. currents C0
Dielectric material
Recovery voltage
, f (t )
K-K+Fourier
(
C0 Capacitance & loss Frequency domain Figure 4.24 Theoretical relations between time domain methods, polarisation and depolarisation currents and recovery voltage, and frequency domain, capacitance and loss measurements.
The dielectric response function, f(t), and the conductivity, , can be determined from polarisation/depolarisation currents if the geometrical capacitance, C0, is known see chapters 4.2.1 and 4.2.2. The arrow is pointing in both directions indicating that the inverse is also true. Recovery voltage can be calculated from the high-frequency component of the relative permittivity, , dielectric response function, f(t), and conductivity, , see chapter 4.3 and Appendix D. The inverse problem, to try to estimate conductivity and dielectric response function from recovery voltage measurements is difficult due to the re-polarisation that occurs in a measurement [36], [37]. The geometrical capacitance, C0, does not affect the recovery voltage. From capacitance and loss (tan) it is easy to calculate the complex relative permittivity, (), when the geometry, C0, is known. The link between time and frequency domain is the Fourier transform and the KramersKronig relations. With these relations, it is possible to transform measured data either from time or frequency domain. The numerical calculation of the Fourier transform can be performed in several ways. Because in the time domain very long measurement periods are needed (tstart to tstop) and because of the non-periodic behaviour of the measured data it is not convenient to use standard FFT algorithms. Since all measurements, both in time and frequency domains are made over a finite interval of time or frequency it is necessary to estimate the measurement data outside this interval. In the time domain, it is more difficult to estimate the short time behaviour (t<tstart) than the long time behaviour (t>tstop) of the insulation material. This is because the short time data is often several magnitudes larger than the long time data resulting in a bigger influence in the Fourier transformation integral. It becomes especially difficult to estimate the short time data if the slope of the measured data is steep like for the XLPE cable. Three different possibilities to transform data will be discussed here, the Hamon approximation [51], a method based on fitting cubic splines to the measurement data before 57
Chapter 4
Fourier transforming [78] and another method also based on curve fitting but here with base functions that have an analytic Fourier transform [38].
4.6.1
The Hamon approximation
It was shown in equation (4.5) that the depolarisation current from a step-voltage excitation of the test object, neglecting the contribution from the delta function, could be written as
i dpol (t ) = (f (t ) f (t + t 1 )) C 0 U 0 < 0 for 0 < t <
(A )
(4.29)
where C0 is the geometric capacitance of the test object, t1 is the charging time and U0 is the applied step-voltage. Assume now that the dielectric response function is a monotonically decreasing function of time. If the test object is charged for a sufficiently long time, the dielectric response function can be expressed as f (t ) = i dpol (t ) C0 U 0 + f (t + t1 ) i dpol (t ) C0 U 0 for t > 0
(1 s )
(4.30)
If now the measured depolarisation current can be piecewise approximated with a Curie-von Schweidler model, see equation (3.36) [51]
i dpol (t ) A t -n
t>0
(4.31)
The Fourier transform of the dielectric response function in equation (4.30) with the time dependence as in equation (4.31) can be expressed as [1] ( n ) = A (1 n ) sin C0 U 0 n i cos 2 2
n -1
0 < n <1
(4.32)
Now only the imaginary part of the complex relative permittivity will be considered according to the Hamon approximation [31]. The imaginary part can be expressed in terms of the depolarisation current idpol(t=t1) at a certain time t=t1 as (
)=
i dpol (t = t 1 ) &0 U 0
0<n<2
(4.33)
provided that and t1 are related to each other as follows n W1 = (1 n ) cos 2

1 n
0.63 for 0.3 < n < 1.2
(4.34)
The expression in equation (4.34) is almost independent of n in the range 0.3<n<1.2 and can with an accuracy of 3% be approximated with the mean value 0.63. The chosen range, 0.3<n<1.2, covers well the experimentally found values of the measured depolarisation currents for most dielectric materials. The Hamon approximation of the imaginary part of the complex electric susceptibility can then be written as [31] (
i dpol (0.63 &0 U 0
i dpol (0.1 f ) 2 C0 U 0f
(4.35)
This approximation is very useful for fast calculations of frequency domain data from time domain data when the above mentioned conditions are valid. 58
Chapter 4
In Figure 4.25, the Hamon approximation has been used to calculate the complex electric susceptibility for the oil/paper insulated HVDC cable. The measured depolarisation current was both compensated, (A), and not compensated, (B), for the finite charging period. It is seen that the calculated imaginary part of the complex electric susceptibility fits well to the measured loss. The difference in result from compensating for the finite charging period or not is seen for the lower frequencies where the compensated curve starts to increase again after the plateau.
101 (A)
C0 = 144 pF U
peak
Hamon
= 3 kV
10 Loss, "
(B)
10
-1
"
10
-2
Measured
10
-6
10
-4
10
-2
10
10
Frequency [Hz]
Figure 4.25
The Hamon approximation is used here to calculate the imaginary part of the complex electric susceptibility for the oil/paper insulated HVDC cable using the depolarisation current which was compensated, (A) and not compensated, (B) for the finite charging period.
In Figure 4.26, the Hamon approximation has been used to calculate the complex electric susceptibility for the resin-rich insulated stator bar. In (A) the depolarisation current was compensated and in (B) it was not compensated for the finite charging period.
100 C = 130 pF
0
(A) Loss, "
Upeak = 0.2 kV
10
-1
Hamon
(B)
"
Measured
10
-2
10
-6
10
-4
10
-2
10
10
Frequency [Hz]
Figure 4.26
The Hamon approximation is here used to calculate the complex electric susceptibility for the resin-rich mica tape insulated stator bar using the depolarisation current which was compensated, (A) and not compensated, (B) for the finite charging period. 59
Chapter 4
The same behaviour is seen for the stator bar as for the HVDC cable, the loss part fits well but there is a difference between compensated (A) and not compensated (B) depolarisation currents. It is well illustrated here the importance of compensating the depolarisation current for the Hamon approximation since there is otherwise a risk of interpreting a false loss peak for the lower frequencies. There is also a decrease in the calculated data for the higher frequencies, which is not seen in the frequency domain. In Figure 4.27, the result from the Hamon approximation is plotted against measured loss for the XLPE insulated AC power cable. The noise in the time domain depolarisation current, see Figure 4.4, is directly reflected in the calculated loss. It is well illustrated here that a low loss material like XLPE is preferable measured in the frequency domain and that a high voltage time domain measurement does not add any more information due to the noise influence. The increase in loss for frequencies lower than 0.1 mHz seen from the time domain measurement is most likely artificial (noise and offset current).
10-2 C = 814 pF
0
Upeak = 3 kV 10 Loss, "

-3
10
-4
Hamon
"
Measured
10
-5
10
-6
10
-4
10
-2
10
10
Frequency [Hz]
Figure 4.27
The Hamon approximation is here used to calculate the complex electric susceptibility for the XLPE insulated AC power. The noise from the depolarisation current is directly reflected in the calculated complex electric susceptibility. The increase in loss seen from the time domain data for lower frequencies than 0.1 mHz is probably artificial.
The general conclusion is that the Hamon approximation works well as a fast estimation of the loss level in a dielectric material from a depolarisation current measurement that was compensated for the finite charging period. It is important to use compensated data, otherwise false loss peaks might appear.
4.6.2
Numerical Fourier transform
Calculating the Fourier transform of a measured dielectric response function f(t) is in most cases very difficult. Since the dielectric response function, for solid dielectric materials often extends over several decades in time it is not possible to use numerical methods that are based on linearly spaced data in time. It is, for example, not practical to use the FFT (Fast Fourier Transform) algorithm [93]. A better method is to fit several cubic splines to the measured logspaced data [38], [78], [79]. These cubic splines will generate a continuous approximation (maintained through the second derivative) to the measured data. The cubic splines can then 60
Chapter 4
be analytically Fourier transformed. A description of the numerical routine used in this chapter is found in Appendix C. Start with the Fourier transform of the dielectric response function
( f
) = f (t ) e -i
dt = f (t ) e -i t dt
0
(4.36)
Since it is experimentally impossible to measure the dielectric response function for all times t > 0, the Fourier integral has to be divided into three parts
( f
) = f (t ) e
t start
dt + f (t ) e -i t dt + f (t ) e -i t dt 0 4 start end 1 4 244 3 t1 4 4 244 3 t1 4 4 244 3

-i t I: Start approximation II: Measured data
t end
(4.37)
III: End approximation
where first and third parts represent times where there exist no measured values of f(t). Therefore the result from the Fourier transform will only be valid in the following frequency region
1 t end < < 1 t start
(4.38)
The two parts, integral (I) and (III), have to be approximated in such a way that they contribute with as small errors as possible see Figure 4.28.
Start approximation Richardson extrapolation At3+Bt2+Ct+D
Dielectric response function f(t)
At-n
Cubic splines
?
Measured data End approximation
A+Be-t/
?
tstart tend Time
Figure 4.28
The dielectric response function f(t) is estimated from a depolarisation current measurement. A number of cubic splines per decade in time are fit to the response function together with a start and end approximation resulting in a Fourier transform that can be analytically solved.
To solve integral (I) the dielectric response function f(t) was approximated with a Curie-von Schweidler model (t-n 0<n<1) or a 3rd order polynomial that was fit to the first decade of the measured data [6]. Another alternative was to use Richardson extrapolation to extrapolate the measured data to t=0 and then fit cubic splines to the set of data from t=0 to t=t2 [69]. To approximate the dielectric response function for times t > tend is not as critical as for t < tstart because most response functions are rapidly decreasing with time. An exponentially decaying 61
Chapter 4
function (A+Bexp(-t/)), was used to make it possible to solve the integral (III) analytically so that the coefficients A, B and were determined [78]. The approach of fitting a number of cubic splines per decade of measured data does not only result in a Fourier transform that is possible to solve analytically, it is also a way of filtering measured data. This approach works well for both the Fourier and inverse Fourier transforms with the only difference that the start and end approximations in the frequency domain are equally important since if the time domain shows strong time dependence, the frequency domain will not.
4.6.2.1
Transformation from time to frequency domain
Here the time domain measured depolarisation currents from the oil/paper insulated cable, resin-rich mica tape insulated stator bar and XLPE insulated AC power cable are used to calculate their dielectric response functions. Each response function is compensated for the finite charging period and then Fourier transformed using the above explained numerical routine. In Figure 4.29, the results from the oil/paper HVDC cable are shown, (A) the measured real part of the complex electric susceptibility compared to the calculated and (B) the measured imaginary part of the complex electric susceptibility compared to calculated. Both the time domain and the frequency domain measurements are made with 3 kV. Calculations have been made with and without (none) start and end approximations.
102 Real part of the susceptibility 10
1
10 Imaginary part of the susceptibility (A)

Spline
DC conductivity 10
1
(B)
=5 10-16 S/m
100 n=-0.95 10-1

-2
100 10-1
-2
Spline
n=-0.90 n=-0.80
n=-0.95 n=-0.90 n=-0.80 n=-0.34
" "
10
Richardson n=-0.34 None

-6
10
None 10
-3
10
-3
3rd order polynom 10

0
Richardson 3rd order polynom 100 102
10
10
-4
10
-2
10
10-6
10-4
10-2
Frequency [Hz]
Frequency [Hz]
Figure 4.29
Comparison between complex electric susceptibility and numerically calculated Fourier transforms of the dielectric response function for the oil/paper insulated HVDC cable. Best fit is found from a start approximation using a Curie-von Schweidler model with n=-0.90 whereas measured time domain data has a slope of n=-0.34.
Using a start approximation which is a 3rd order polynomial or Richardson extrapolation works well for the loss part but gives a real part which is too low for frequencies in the range of 1 mHz to 0.1 Hz. The Curie-von Schweidler model with n=-0.34, the initial slope of the measured dielectric response function, gives also a good estimation for the loss part but too low values for the real part. The value of n is then increased, n=-0.80, -0.90, -0.95, showing a corresponding increase in the real part but no change in the loss part. This illustrates the 62
Chapter 4
importance of a correct estimation of the dielectric response function for times shorter that t=tstart, start of measurement. The real part of the complex electric susceptibility is more sensitive to the start approximation than the imaginary part which can be understood since the real part corresponds to the cosines transform (cos(x)1 when x0) and the imaginary part corresponds to the sinus transform (sin(x)0 when x0) see equation (3.25). The loss contribution from the conductivity is also plotted in (B) showing the relation between the pure DC conduction loss and the measured loss. To determine the conductivity from a frequency domain measurement with a good accuracy it is needed to measure at frequencies around 0.01 mHz to 0.1 mHz. In Figure 4.30, results from the numerical Fourier transforms of the measured dielectric response function for the resin-rich insulated stator bar are compared with measured complex electric susceptibility. The measurement in the time domain was made with 3 kV and the measurement in the frequency domain was made with 200V peak. The start approximation is important also for the stator bar, especially for the real part of the complex electric susceptibility. As start approximation the Curie-von Schweidler model with n around -0.9 works best both for the real and imaginary parts. This should be compared to the initial slope of the measured dielectric response function, n=-0.80, which will give a good estimation for the loss part but too low values for the real part. The calculated losses have a tendency to start to decrease earlier than the measured values. Worth to mention regarding this is that the measurement electrode on the stator bar was a semi conductive tape, which might have had voltage dependence. There is no indication of conduction current, loss curve slope of -1 in log-log scale, in either the measured or the calculated data.
100 Real part of the susceptibility n=-0.95 Imaginary part of the susceptibility (A) n=-0.90 100 (B)
10
-1
Spline
-
n=-0.80 None Richardson
10
-1
Spline
n=-0.95 n=-0.90 n=-0.80
" "
None 10
-2
3rd order polynom 10

-2
Richardson 3rd order polynom 100 102
10-6
10-4
10-2
100
102
10-6
10-4
10-2
Frequency [Hz]
Frequency [Hz]
Figure 4.30
Comparison between complex electric susceptibility and numerically calculated Fourier transforms of the dielectric response function for the resinrich mica tape insulated stator bar. Best fit is found from a start approximation using a Curie-von Schweidler model with n=-0.90 whereas measured time domain data has a slope of n=-0.80.
In Figure 4.31 results from the numerical Fourier transforms of the measured dielectric response function for the XLPE insulated AC power cable are compared with measured complex electric susceptibility. The measurements in time and frequency domains were made with 3 kV. It is seen that none of the start approximations used here will give a good fit to both the measured real and imaginary part of the complex susceptibility. This can be understood from looking at the measured dielectric response function in Figure 4.4 showing 63
Chapter 4
that the slope in the beginning of the measurement is steeper than t-1. Such rapid time dependence is not possible to model with the Curie-von Schweidler model (will give singular integral). For the start approximations using the 3rd order polynomial and the Richardson extrapolation scheme a higher number of splines per decade of data was chosen resulting in more noise which illustrates the noise suppressing feature of the spline fit. The calculated imaginary part of the complex electric susceptibility shows that the losses are increasing below 0.1 mHz. This is a direct effect related to the problem with the offset current and the noise level in the cable shown in Figure 4.4. Like the Hamon approximation it seems to be difficult to measure low loss materials in the time domain and then successfully Fourier transform those results to the frequency domain, especially for the lower frequencies.
10-1 Real part of the susceptibility 10-1
-
n=-0.99
10
-2
Imaginary part of the susceptibility
(A)
(B) 10-2 n=-0.95 n=-0.90 n=-0.80
10
-3
Spline
" "
n=-0.80 Richardson 3rd order polynom n=-0.34 None
10
-3
10-4
Spline
10-4
Richardson 3rd order polynom None 10-4 10-2 100 102
10
-5
10-6
10-4
10-2
100
102
10
-5
10-6
Frequency [Hz]
Frequency [Hz]
Figure 4.31
Comparison between complex electric susceptibility and numerically calculated Fourier transforms of the dielectric response function for the XLPE insulated AC power cable. No good start approximation can be found since the measured time domain data has a slope n=-1.19 making the start integral singular.
4.6.2.2
Transformation from frequency to time domain
Here the frequency domain measurements from the oil/paper insulated cable, resin-rich mica tape insulated stator bar and XLPE insulated AC power cable are used to calculate their inverse Fourier transforms. The inverse Fourier transform is calculated using the above explained numerical routine, the start and end approximations in the frequency domain are equally important since if the time domain shows strong time dependence, the frequency domain will not. Interesting here is also to compare the two transforms, first using the real part and then the imaginary part of the measured complex electric susceptibility. The two transforms should give the same result and if not then it is most likely that the real part is shifted because the wrong high-frequency component of the relative permittivity, , was subtracted or the imaginary part shows strong conductivity, . In Figure 4.32, the results are shown from the inverse Fourier transform of the measured real part, (A), and imaginary part, (B), of the complex electric susceptibility for the oil/paper insulated HVDC cable. It can be seen that when using no start or end approximations (none) the result will be very noisy. If using a 3rd order polynomial or Richardson extrapolation as a start approximation, it makes no difference in (A) or (B). The best start approximation is 64
Chapter 4
found using the Curie-von Schweidler model with (A) n=-0.68 and (B) n=-0.60 which are the slopes found from the measurement data, first decade of data after the start approximation. A Curie-von Schweidler model was also used as an end approximation to extend the measured data for one extra decade with the same n as found from the last decade of measured data. After this extension an exponentially decaying function, the same as used from time to frequency domain (A+Bexp(-t/)), was added to end the transform.
102 Dielectric response function, f(t) 10 10
1
102 Dielectric response function, f(t) (A) f(t)Spline (B) 10 10

1
f(t)Spline
0
10-1 10
-2
10-1 10
-2
n=-0.60 Richardson 3rd order polynom None
10-3 10 10
-4
None Richardson 3rd order polynom 10-2 100 Time [s] 102
n=-0.68 f(t)
10-3 10 10
-4
f(t)
-5
-5
10-4
104
10-4
10-2
100 Time [s]
102
104
Figure 4.32
Comparison between dielectric response function and numerically calculated inverse Fourier transforms of (A) real part and (B) imaginary part of the complex electric susceptibility for the oil/paper insulated HVDC cable.
In Figure 4.33, the results are shown from the inverse Fourier transform of the measured real part, (A), and imaginary part, (B), of the complex electric susceptibility for the resin-rich mica tape insulated stator bar. It can be seen also for the stator bar that when using no start or end approximations (none) the result will be very noisy.
102 Dielectric response function, f(t) 10
1
102 Dielectric response function, f(t) f(t)

Spline
(A)
10
f(t)
Spline
(B)
100 10
-1
100 10
-1
10-2 None 10
-3
10-2 10
-3
None
n=-0.11 Richardson 3rd order polynom
n=-0.09 Richardson f(t)
10-4 10-5 3rd order polynom 10

-6
10-4 10-5 10
-6
f(t)
10
-4
10
-2
10
10
10
10
-4
10
-2
10
10
10
Time [s]
Time [s]
Figure 4.33
Comparison between dielectric response function and numerically calculated inverse Fourier transforms of (A) real part and (B) imaginary part of the complex electric susceptibility for the resin-rich mica tape insulated stator bar.
Best start approximations are found using the Curie-von Schweidler model with (A) n=-0.09 and (B) n=-0.11. The Curie-von Schweidler model was also used as an end approximation to 65
Chapter 4
extend the measured data for one extra decade, then using the same exponentially decaying function as above to end the transform. Looking carefully at the results in (A) and (B) it can be seen that the transformed frequency domain response gives a dielectric response function which is a bit larger in magnitude but correct in shape compared with the measured one in the time domain. In Figure 4.34, the results are shown from the inverse Fourier transform of the measured real part, (A), and imaginary part, (B), of the complex electric susceptibility for the XLPE insulated AC power cable. Without any start or end approximation the result is not only noisy but it is also shifted by a factor towards longer times. All three types of start and end approximations used here work well both for the real and imaginary part of the complex electric susceptibility. There is a better fit between calculated and measured data in (B) compared to (A) where the fit is good up until one second after which there is a plateau in the calculated data. The deviation seen in (A) around one second is from the increase in the measured real part of the complex electric susceptibility seen in Figure 4.22 for the lower frequencies. This is probably a measurement error because the change in real part has no corresponding change in the imaginary part.
-2
(A) f(t)Spline None n=-0.11 Richardson 3rd order polynom Dielectric response function, f(t)
102 100 10
-2
(B) f(t)Spline
10-4 10
-6
10-4 10
-6
n=0.08 Richardson 3rd order polynom
None
f(t)
10-8 10
-10
f(t)
10-8 10
-10
10
-4
10
-2
10
10
10
10-4
10-2
100 Time [s]
102
104
Time [s]
Figure 4.34
Comparison between dielectric response function and numerically calculated inverse Fourier transforms of (A) real part and (B) imaginary part of the complex electric susceptibility for the XLPE insulated AC power cable.
4.6.3
Fourier transform using analytic base functions
Another method to calculate both the Fourier transform and the inverse Fourier transform is to fit an expression to either time or frequency domain data that has an analytic transformation. A commonly used method is to expand the measured data as a sum of exponentials, Debye models with different time constants i which are logarithmically spaced in time, see equation (3.35) [30], [51]. This often works well, but using Debye models for the start and end approximations, see Figure 4.28, can sometimes generate errors. For solid dielectric materials, the dielectric response function can often be piecewise approximated with the Curie-von Schweidler model. Combining the analytic features of the Debye model with the often experimentally observed Curie-von Schweidler model the dielectric response function f(t) can successfully be approximated as 66
Chapter 4
t b t ms t b0s f (t ) = a s e 0s b + 1 e s =0 0s
t b 0s
ns
0 < ms < 1 0 < ns < 2
(4.39)
where ms is the slope in a log-log scale before the breakpoint t=b0s and ns is the slope after. The scaling with b0s is important for smooth transitions between the slopes, especially when there is a big change in slope. This expression has the same asymptotic behaviour as the general response expression in equation (3.37) but it has an analytic Fourier transform [51], [54]. The complex electric susceptibility can be written as [56], [90]
(
ms ns ns b 0s (1 m s ) (1 n s ) (1 n s ) b 0s b 0s ) = as + 1 ms 1 n s 1 n s ( ) ( ) ( ) + + i 1 b 1 b i i s =0 0s 0s
0 < ms < 1 0 < ns < 2
(4.40)
Often it is enough to use only two base functions for f(t) or (). The start and end approximations are adjusted by changing ms in the fist base function and ns in the last base function. It is important to notice that the expressions in equations (4.39) and (4.40) are only curve fittings and have no real physical significance. Nevertheless, they provide a very fast and simple engineering tool for transforming data from either time or frequency domain. In Figure 4.35 (A) two examples of the base functions in equation (4.39) are plotted against time and in (B) their corresponding Fourier transforms given by equation (4.40) are split in real and imaginary parts and plotted against frequency. By changing the slopes ms and ns in the time domain the shape of the peak in the frequency domain is changed accordingly. From a practical point of view it often turns out to be easier to work with the base functions in the frequency domain since it is easier to fit data to a peak than to a monotonically decreasing function like the dielectric response function in the time domain.
m = 0.8 log(Dielectric response function, f(t))
1
(A) f(t)1 log(Complex susceptibility,*())
(B) 1 -1 + n2 -1 + m
2 1
m = 0.2
2
f(t)2 n2 = 1.2 n = 1.8

1
-1 + m -1 + n1 1 = 1/b 2
t=b
1,2
log(Time)
1,2
log(Freqency)
Figure 4.35
Example of an analytic base function, which has the same asymptotic behaviour as the general response function. (A) time domain and (B) corresponding transformed function in frequency domain, (B).
In Figure 4.36 (A), the result is shown from fitting two base functions in the time domain to the measured dielectric response function for the oil/paper insulated HVDC cable. The measured response function is compensated for the finite charging period. The first base function was put with the breakpoint at b01=2000 seconds with slopes m1=0.35 and n1=0.8 and the second base function was put with the breakpoint at b02=10-5 seconds with slopes m2=0.4 and n2=1.12. In (B) the result from the Fourier transform is compared in the frequency domain 67
Chapter 4
with measured real and imaginary part of the complex susceptibility. The fit to the measured data in both time and frequency domains are good showing that the received information from the two measurements, polarisation and depolarisation currents in the time domain and real and imaginary parts of the complex susceptibility in the frequency domain, are equivalent.
0
-1
f(t)Analytic = f1(t) + f2(t)
(A)
Parameters TD : C 0 = 144 pF U 0 = 3 kV
-2
10-3 10
-4
f(t)Analytic = f1(t)
f(t)
Measured
m1 = 0.35 n 1 = 0 .8 b 01 = 2000
10-5 10
-6 -2 0 2 4 6
10
10
10
10 10
1
10
Time [s] Complex electric susceptibility
Parameters FD : C 0 = 144 pF U peak = 3 kV m 2 = 0 .4 n 2 = 1.12 b 02 = 10 5
(B) 10
0
-
10
-1
1 + 2
" "
10
-2
Analytic
Analytic = 1 10
-3
1+ 2 10 10 10 Frequency [Hz]
-2 0 2
10
-6
10
-4
10
10
Figure 4.36
(A) two base functions which are fit to the measured dielectric response function for the oil/paper insulated HVDC cable. (B) the same two base functions transformed to the frequency domain and compared with measured data.
In Figure 4.37 (B), the result is shown from fitting three base functions in the frequency domain to the measured real and imaginary part of the complex susceptibility for the resinrich mica tape insulated stator bar. The first base function was put at b01=310-2 seconds with slopes m1=0.8 and n1=0.96, the second at b02=2104 seconds with m2=0.4 and n2=1.15 and the third at b03=107 seconds with m3=0.4 and n3=1.12. It was chosen here to use three base functions due to the shape of the measured loss curve, but a reasonable good result would have been reached with only two base functions. In (A), the analytic dielectric response function, plotted with the same parameters as for the three base functions in the frequency domain, is compared with the measured response function. The measured dielectric response function in (A) is compensated for the finite charging period.
68

10 Dielectric response function, f(t) 10
0
Chapter 4
(A)
-1
Parameters TD : C 0 = 130 pF U 0 = 3 kV m 3 = 0 .4 n 3 = 1.12
10-2 10 10
-3 -4
f(t)
Measured
10-5 10 10
-6 -7
f(t)
Analytic
= f (t) + f (t)
1 2
b 03 = 10 7
f(t)Analytic = f1(t) + f2(t) + f3(t)

-2
10
10
10
10 10 Complex electric susceptibility

1
10
Time [s]
Parameters FD : C 0 = 144 pF U peak = 0.2 kV m 1 = 0 .8 n 1 = 0.96 b 01 = 3 10

2
+ +
1 2
(B)
Analytic
10
= +
1
m 2 = 0 .4 n 2 = 1.15 b 02 = 2 10
4
10
-1
10-2
" "
1+ 2 + 3 Analytic = 1 + 2
-3
10-3 -7 10
10
-5
10
10 10 Frequency [Hz]
-1
10
10
Figure 4.37
(A) measured dielectric response function compared with calculated. (B) three base functions are fit to the measured real and imaginary part of the complex susceptibility for the resin-rich mica tape insulated stator bar. The same parameters are used to calculate the time domain response.
In Figure 4.38 (A), the result is shown from fitting two base functions in the time domain to the measured dielectric response function for the XLPE insulated AC power cable. The first base function was put at b01=10-6 seconds with slopes m1=0.8 and n1=1.21 and the second one at b02=107 seconds with slopes m2=0.8 and n2=1.24. In (B) the result from the Fourier transform is compared in the frequency domain with measured real and imaginary part of the complex susceptibility. The increase in loss seen from the fit in the time domain can be questioned in a similar way as with the spline approach used above. These analytic base functions provide a simple tool for transforming data between time and frequency domains. They also give a good illustration of how, by changing the slopes in either domain, the Fourier transform connects the two domains. As seen from the results in this chapter the analytic base functions also provide a simple way of describing a dielectric material over a wide range of frequencies. Base functions fitted to measured time domain data will give the low frequency part and base functions fitted to measured frequency domain data will give the high frequency part.
69
Chapter 4
Dielectric response function, f(t) (A) f(t)
Analytic
10-2
-4
Parameters TD : C 0 = 814 pF U 0 = 3 kV m 1 = 0 .8 m 2 = 0 .8 n 2 = 1.24 b 02 = 10 7
= f (t) + f (t)
1 2
10
10
-6
f(t)
Measured
n 1 = 1.21 b 01 = 10
6
10
-8
10
-10
10
-2
10
10
10 10 Complex electric susceptibility

-1
10
Time [s]
Parameters FD : C 0 = 814 pF U peak = 3 kV
Analytic
= +
1
(B)
10
-2
10
-3
10-4
Analytic
= +
1
" "
10-5 -7 10
10
-5
10
-3
-1
10
10
Figure 4.38
(A) two base functions which are fit to the measured dielectric response function for the XLPE insulated AC power cable. (B) the same two base functions transformed to the frequency domain and compared with measured data.
4.6.4
Numerical calculation of the Kramers-Kronig relations
The Kramers-Kronig relations are relating the real and imaginary part of the complex electric susceptibility to each other. Recalling equation (3.34), these relations can be written as
( ( x ) = 2 lim 2 a
a 0
(x) dx x 2
a
(4.41)
)= 2
lim
a 0
(x) dx x 2
2
In a frequency domain measurement, the complex relative permittivity is measured and it is related to the complex electric susceptibility in the following way
70
Chapter 4
)= 1 + ( ) i 4 24 3
=
+ ( ) 0 14 4 244 3 =
(4.42)
where is the conductivity and is the high-frequency component of the relative permittivity. The value of represents the amount of polarisation at higher frequencies, outside the measurement window where the complex relative permittivity is measured. For most dielectric material at room temperature and at reasonable high frequencies, the conductivity can be neglected. This means that the measured loss part, "(), is proportional to the imaginary part of the complex electric susceptibility, "(). With these assumptions, the real part of the complex electric susceptibility, (), can be calculated using the KramersKronig relations. From the real part of the measured complex relative permittivity, (), a value estimating can be subtracted so that the slope and magnitude correlate to the calculated (). A practical problem here is that the loss part is measured in a finite, often narrow, frequency window. It is therefore important to make good start and end approximations to the measured data and depending on how they are done the value found for will be moving with frequency. Most likely, the found is the value at a higher frequency than the highest in the measurement window. The easiest case to make start and end approximations is when a distinct loss peak is measured. The calculated () will then represent the change in () related to that specific loss mechanism and the interpretation of will be more straightforward. If the dielectric response function f(t) is finite for all values of t0, zero for t<0 and f(t)0 as t, see equations (3.29) to (3.33), then the Fourier transform of this f(t) will give a complex electric susceptibility which has a real and imaginary part that will fulfil Kramers-Kronig relations [64], [66]. An example of such a dielectric response function is the proposed sum of analytic base functions found in equation (4.39). Therefore, the real and imaginary part of the complex electric susceptibility in equation (4.40) will always fulfil Kramers-Kronigs relations. If then the imaginary part in equation (4.40) is fit to the measured loss, with the conductivity subtracted if influencing data, then the real part is directly given from the same set of parameters. This is one way to calculate the Kramers-Kronig relations. Another way to numerically calculate the Kramers-Kronig relations are used here, which is based on the same spline fitting and integration method, used for the numerical Fourier transform in chapter 4.6.2, see also Appendix C. The first step in this numerical approach is to rewrite the equations in (4.41) as
( ( x ) = 2 lim 2 a
a 0 a (x) (x) 1 (x) = + dx lim dx 2 a x x x+ 0 a
(4.43)
)= 2
lim
a 0
(x) (x) 1 (x) dx = lim dx 2 a + x x x 0

a 2
This gave four integrals that have to be solved, but the same principal can be used to solve all of them so it is enough to study one of them. () and () are here replaced with a function g() that has the same principal features. Study now
71
Chapter 4
I1 (
)=
0
g(x) dx = x
Start approximation
2 g(x) g(x) g(x) dx dx + dx + x x x 0 1 2 1 4 24 3 1 4 24 3 1 4 24 3
(4.44)
Spline approximation
End approximation
where g(x) is expressed as a sum of cubic splines that are fit to the measured data for 1<< 2, see Appendix C. Outside the region of measured data a good start and end approximation must be found. The integrals in equation (4.44) have a Cauchy-type of singularity that can be evaluated under the assumption that g(x) is a smooth function in the sense that its first and second derivative is finite in the integration interval [6]. The idea is to integrate by parts two times giving the following result
I2 (
b part. g (b ) ln b part. g(x) = dx = ln x g(x)dx = = int ( ) int x g a ln a a a g (b ) ln b g (b ) (ln b 1) (b ) b = + (ln x 1) (x ) g (x)dx g (a ) ln a g (a ) (ln a 1) (a ) a
)=
(4.45)
This expression is well behaved especially since g(x) in this case is a cubic spline that is fit to the measured data. The summation over N splines in the region 1<< 2 together with solving the last integral in equation (4.45) can be done analytically with some patience and help of Maple V [72]. The start approximation can be done in the same way as for the Fourier transform using a 3rd order polynomial or Richardson extrapolation to estimate data points for < 1 which are then used to fit splines to between =0 and = 1. It is also possible to use a Curie-von Schweidler model giving the following start approximation [1]
I Start 1 (
) = g(x) x
Start approximation
dx 0 1 4 24 3 1 + 1 2n
-n
2 x x -n 1 1 -n x x dx 1 x 1 = + + + Ldx = = < x 0 0
1
-n = 1 1 n
1 + 1 3 n
-n
3 -n 1 = 1 + L s =1 s n
n <1 >
1
(4.46)
Note that the Curie-von Schweidler model is not well behaved in the same way as the cubic splines are above, especially for x0. This approximation works only for n<1 and >1. The end approximation is as important as the start approximation and is also approximated with a Curie-von Schweidler model if the slope of the measured data is negative in a log-log scale, n<0. If n>0 then the data is extended for one extra decade with the Curie-von Schweidler model and then approximated with an exponential decaying function. Such an end approximation can be written as
I End 1 (
) = g(x) x
2
dx
End approximation
1 4 24 3
s = x e -x =e x dx = = ds dx 2
e -s ds = s
x =s = =e dx = ds
e dx = e x
-x
Ei
(4.47)
<
2
where Ei(x) is the exponential integral function [1], [90]. 72
Chapter 4
The total result is then expressed when summing up all four integrals in equation (4.43). This method works reasonable well but the start and end approximations influence the result sometimes too much. There are also other possible numerical methods to use, which are based on modified Simpson rules or different types of Fourier series [4], [5], [50]. In Figure 4.39 the imaginary part (solid line) of the sum of two analytically Fourier transformed base functions, see equation (4.40), have been fit to the measured loss part of the oil/paper insulated HVDC cable. From the measured real part of the complex relative permittivity, (), a constant value was subtracted so that the result would correlate in magnitude and slope with the real part (solid line) of the two base functions. The real part was given directly after fitting the imaginary part to the measured loss part. The high-frequency component of the relative permittivity was estimated to 4.0 for the HVDC cable. It was verified from time domain measurements of polarisation and depolarisation currents that the loss part in this frequency range was not strongly influenced by DC conductivity. The numerical routine, see Appendix F, was first used to calculate the real part of the complex susceptibility (dashed line) from the measured imaginary part, (), and then to calculate the imaginary part (dashed line) from the measured real part, ()- . The Curie-von Schweidler model together with an exponential decaying function was used as both start and end approximations for the numerical routine.
101 Complex susceptibility Analytic Numerical 100 4.0
-
10
-1
10
-2
"
10
-5
10
-3
10
-3
-1
10
Figure 4.39
Calculated Kramers-Kronig relations, solid line the fit of two analytic base functions and dashed line numerically calculated, compared to measured real and imaginary part of the complex susceptibility for the oil/paper insulated HVDC cable.
In Figure 4.40 the same calculations, but with three base functions, have been done for the resin-rich mica tape insulated stator bar as were done for the HVDC cable above. The highfrequency component of the relative permittivity was estimated to 4.3 for the stator bar. The calculated real part of the complex susceptibility from the numerical routine is probably decreasing too fast for the higher frequencies as a result of an inaccurate end approximation of the measured loss.
73
Chapter 4
100 Analytic Numerical Complex susceptibility

4.3
-
10
-1
"
10
-2
10-5
10-3
10-1 101 Frequency [Hz]
103
Figure 4.40
Calculated Kramers-Kronig relations, solid line the fit of two analytic base functions and dashed line numerically calculated, compared to measured real and imaginary part of the complex susceptibility for the resin-rich mica tape insulated stator bar.
In Figure 4.41 the same calculations as for the HVDC cable have been done but here for the XLPE insulated AC power cable. The high-frequency component of the relative permittivity for the XLPE cable was estimated to be 2.3, based only on the result from the fit of the two analytic base functions. The calculated real part of the complex electric susceptibility from the numerical routine is also here influenced by a poor end approximation of the measured loss. It seams to be easier and more flexible to make good end approximations with the analytically Fourier transformed base functions. The numerically calculated loss part shows a strong increase in loss for the lower frequencies, which is a result from an increase in the measured real part of the complex electric susceptibility. This increase is probably, as discussed earlier, a measurement error.
10-1 Analytic Numerical 2.3
Complex susceptibility
10
-2
10
-3
"
10
-4
10-5
10-3
10-1
Frequency [Hz]
101
103
Figure 4.41
Calculated Kramers-Kronig relations, solid line the fit of two analytic base functions and dashed line numerically calculated, compared to measured real and imaginary part of the complex susceptibility for the XLPE insulated AC power cable.
74
Chapter 4
The Kramers-Kronig relations were here used to estimate the high-frequency component of the relative permittivity, , for measurement in a rather narrow frequency window. Doing several measurements at different temperatures, which might also give a possibility to see the influence of the DC conductivity will extend the frequency window.
4.7
Summary
Three different diagnostic methods, polarisation/depolarisation currents, recovery voltage and loss and capacitance applied to three different insulation systems (oil/paper, resin-rich mica tape and XLPE) with different characteristics have been studied in this chapter. For measurements of polarisation and depolarisation currents, it was shown the importance of compensating for the finite charging period when estimating the dielectric response function and conductivity. This is important regardless of insulation system. The forward and inverse problems were discussed for the recovery voltage method. The forward problem to calculate the recovery voltage given f(t), and is straightforward as long as the grounding time, t2-t1, is well defined. The inverse problem is more difficult but when the general behaviour of the dielectric response function is known it can be calculated by minimising the current to an analytic expression. An example of this was given together with an attempt to estimate also the conductivity from a recovery voltage measurement. The problem with offset currents and memory effects from previous measurements are also discussed and illustrated. Frequency domain measurements of capacitance and loss as a function of frequency are shown for all three insulation systems. Different physical mechanisms can sometimes be easier to identify in the frequency domain than in the time domain. For the oil/paper insulation system conductivity is easily separated from a LFD mechanism by plotting the data as complex susceptibility. Looking at the results from the three different diagnostic methods used on the three different insulation systems a few conclusions can be made. The best way is to choose diagnostic method and domain according to the degree of dielectric loss in the system and according to the time/frequency range a certain interesting phenomena is expected to be found in. Different methods to Fourier transform data between time and frequency domains have been studied. The Hamon approximation works well and is fast but it is important to compensate data for the finite charging time which otherwise will appear as a peak in the frequency domain. A numerical Fourier transform algorithm based on a spline approximation was also studied. Extending data outside the measured interval is important for the transform but also difficult. An example was when transforming data from the time to the frequency domain for the XLPE cable when time domain data had a steeper slope than t-1 for short times. Analytic base functions that were fit to measured data in both the time and the frequency domain have been studied. The advantage with these functions is that they have an analytic Fourier transform and fulfils Kramers-Kronig relations directly resulting in fast calculations. To use analytic base functions is a simple method based on no physics (curve fit) but it gives good flexibility especially since start and end approximation can be easily changed. If the user 75
Chapter 4
also has some extra knowledge about the insulation system outside the measured data, it can just be added with an extra base function.
76
Test cell for processing resin-rich insulation
Chapter 5
5.1
Introduction
The basic idea behind building a test cell was to simulate the manufacturing process of the electrical insulation in rotating machines. The insulation system, studied here, is based on a pre-impregnated (resin-rich) mica tape that has to be cured under a given temperature and pressure profile in order to reach maximum electrical, thermal and mechanical performance. The test cell was built to match these demands. In this chapter, all the design steps of the test cell are described. The chapter is structured in such a way that the reader can follow the development from simple idea to ready test cell. Since the development of the test cell is based on gathered knowledge throughout the whole project, this chapter might not follow the most logical structure but it describes the progress of work. Initially there were discussions about buying a ready-made test cell and inquiries were made about different options like polymeric presses. However, it was decided to build the test cell within the project mainly because no commercial option fulfilled all our demands but also because of the engineering challenge. Pictures of the test cell are shown in Appendix F.
5.2
Designing the test cell
It was important to be able to make small and easy to handle test samples that still resembled the factory-processed insulation in as many ways as possible. In the first designing step, it was decided to use a simple flat one-dimensional geometry for the samples, see Figure 5.1. F x
Heating plate Pressing force : F=const Top : T=TT (t) Sample Heating plate Bottom : T=TB(t)
Figure 5.1
Basic idea for the design of the test cell. The test sample was heated from two directions with the same temperature path during which the sample is pressed together with a constant force.
77
Chapter 5
This geometry gives the advantage of a homogenous electric field when measuring capacitance and loss and an easy to measure one-dimensional temperature gradient in the sample. The heating plates should be programmable to follow a temperature path with both heating and cooling segments. Heating should take place at a minimum rate of 5C/min whereas cooling should take place at a minimum rate of 10C/min. These are the minimum requirements used in the factory process. The heating plates were made in the workshop from two copper plates, dimensions 17017025 mm3, equipped both with a heating-element and five cooling channels, see Figure 5.2. The heating-elements were lowered into the top of the copper plates using a mill machine. To ensure good heat conduction small copper chips were packed down between the heating-elements and the copper plates. The heating-elements were also shaped in such way that the influence from the large resistive current, giving possible magnetic noise in the measurement, was minimised. To fix everything a thinner copper plate were screwed on top of the heating plates. Five parallel cooling channels for each heating plate were drilled with a diameter of dcc=5 mm. Copper has a specific heat of CP390 J/kgK and density of d8930 kg/m3 at 20C [46]. The Debye temperature for copper is around D315 K, [64], and since the test cell will be operating in the temperature range 10CT200C the specific heat can be seen as constant. The amount of energy needed to raise the temperature T degrees in the test cell can be calculated as
Q = m C p T = V
d
C p T
(J)
(5.1)
For example, raising the temperature one degree, T=1, requires the energy Q=2.5 kJ. To meet the minimum factory requirements, heating 5C/min and cooling 10C/min, each heating-element must generate at least 210 W and each set of cooling channels remove at least 420 W
Top view L = 170 mm Side view L = 170 mm H = 25 mm
L = 170 mm
Cooling channels
Cu
Heating element
dcc=5 mm
Figure 5.2
The layout for the two heating plates. Heating was done electrically and cooling done through cooling channels circulating water.
It is necessary to consider the heat flows to and from the test cell when dimensioning heating power (heating-elements) and cooling power (cooling channels). A temperature difference will cause an energy flow, dP, through the surface, dS, in order to reduce this difference. There are three different modes of heat transfer [47],[48],[57]. The first, heat conduction, is connected to atom (molecular) motion of the material and can be expressed as 78

dP = - dS dT ds (W)
Chapter 5 (5.2)
where (W/mK) is the thermal conductivity and T=T(s) the temperature distribution in sdirection (perpendicular to surface dS) of the material. The second mode of heat transfer is convection, which is energy transfer by motion of a liquid or gas. This can be written as dP = c
dS (Tsurface - Toutside ) 14 4 244 3

T
(W )
(5.3)
where c (W/m2K) is the convection heat transfer coefficient. The temperature difference, T, is found between the surface, dS, and a point some distance out in the moving medium. There are two different types of convection, free and forced. Free convection is for example the motion of air outside the test cell, c will be dependent on T. The cooling channels of the test cell are an example of forced convection, free convection can be neglected and c is not dependent on T. The third mode of heat transfer is radiation, which is energy transferred by electromagnetic radiation [86]. For a grey surface (absorptivity equals emissivity) the net flow of energy can be written as
dP = e
S- B
4 4 ) dS (Tsurface - Toutside
(W )
(5.4)
where e is the emissivity (0 e 1) and S-B=5.670510-8 (W/m2K4) is the Stefan-Boltzmann constant. The total heat flow from the heating plates can be calculated summing up the contributions from equations (5.2), (5.3) and (5.4).
5.2.1
Dimensioning heating
The heat loss of each heating plate was assumed to be the critical parameter for dimensioning heating of the test cell. To calculate the total heat loss from one heating plate with or without any thermal insulation the assumptions are made that the heat conductivity of air outside the test cell is air0 W/mK and emissivity of the surface of the test cell is e1. The area of the test cell surface is Stot=Sup+Sdown+Ssides, Sup=Sdown0.172 m2 and Ssides40.170.025 m2. The convection heat transfer coefficient, c, is approximated for an outside air at standard room temperature, Toutside300 K, and standard pressure with [48]
c c c
2.5 4 (Tsurface - 300) 1.4 4 (Tsurface - 300) 1.4 4 (Tsurface - 300) h
Horizontal surface, up Horizontal surface, down Vertical surface, h = height (m)
(5.5)
The total loss can then be expressed for one heating plate without thermal insulation as
No thermal Ptot = -S tot ( insulation
c
(Tsurface - 300) +
S- B
4 (Tsurface - 300 4 ))
(5.6)
The heating plates were now thermally insulated with a 3 mm thick layer of a mica laminate (Foliemikanit) and two 10 mm thick layers of a fibreglass-polyester laminate (1220), see Figure 5.3. The mica layer has a heat conduction of mica0.2 W/mK and a maximum continuous working temperature of 500C. These features make mica ideal as the first thermal insulation layer. The fibreglass-polyester laminate has a heat conduction of 12200.3 W/mK and a maximum continuous working temperature of 220C. 79
Chapter 5
Temperature
Heating plate Mica Laminate Laminate 1220 1220
Ptot=P+P+P 0
Tsurface
T0 Tsurface
Outside air
Toutside
s 1=0 s 2=3 s 3=13 s 4=23 Distance [mm]
Figure 5.3
Temperature gradient for the thermal insulation of the heating plate. Three layers were used, the first was chosen as mica and the following two as a laminate of fibreglass and polyester (1220).
In Figure 5.3 it is also assumed that there is no heat loss from convection or radiation from the inner surfaces, s1, s2, and s3. The total loss, Ptot, and the temperature on the surface of the test cell, Tsurface, can be calculated from putting equations (5.2), (5.3), (5.4) and (5.5) into a nonlinear system of equations of the following form
1 s 2 - s1 s 4 - s 2 Thermal Ptot = -S tot + ) (Tsurface - Tsurface 1220 mica insulation 4 4 ( = P -S T tot c surface Toutside ) - S tot (Tsurface - Toutside ) tot
(5.7)
The results from calculating the total heat loss from one heating plate with, see equation (5.6), or without, see equation system (5.7), thermal insulation are shown in Figure 5.4 (A).
300
Ptot: No insulation
120
Ptot: Insulation Tsurface, calculated Tsurface, measured T outside, measured
Calculated power loss [W]
250 200 150 100 50 (A) 0 0
100 Temperature [C]
80
60
40 (B) 50 100 150 200 20
50
100
150
200
Temperature heating-plate [C]
Figure 5.4
(A) Calculated heat loss for one heating plate with and without thermal insulation. (B) Calculated temperature on the surface of the thermal insulation of the test cell compared to measured.
It can be seen that there is a considerable difference in calculated power loss between the two cases. To chose the output power for the heating-element the maximum heat loss with thermal insulation, Ploss(T=200C)125W, plus the power needed to meet the rate of factory heating, 80
Chapter 5
Pheat210W (+5C/min), were considered. This resulted in the choice of a heating-element with an output power of 670 W. Such a high output power was chosen since there are a few uncertainties that are not considered here. For example, the heat transfer between the heatingelement and the heating plate (block of copper) and the change in heat capacity for a resinrich test sample during curing. In Figure 5.4 (B) the calculated temperature on the outside surface of the thermal insulation is compared to the measured indicating that there is less power loss than calculated, especially for higher temperatures. See Chapter 5.3 for measured values of the heating rate.
5.2.2
Dimensioning cooling
To calculate the cooling effect from using cooling channels, see Figure 5.2, is very difficult. However, a simple engineering estimation can be done [47], [48]. The fluid flow through the cooling channels is not due to a temperature gradient like in free convection but due to an external force pressing the cooling fluid through the channels. This is called forced convection. For forced convection, the heat conduction is strongly dependent on the state of the fluid. There are two states called laminar flow and turbulent flow. To distinguish between the two, a dimensionless quantity called the Reynolds number was introduced Reynolds number : Re = v
d
d cc
(5.8)
where v is the mean fluid velocity, d the fluid density, dcc the diameter of the cooling channels and the dynamic viscosity. The transition between laminar and turbulent flow does not occur at a specific value but gradually in the range from around 2000 to around 5000. Assume now that the cooling liquid is water at a temperature of 7C and that the mean fluid velocity v12 m/s. The mean fluid velocity can be estimated by measuring the volume flow per second and then scaling this with the known cross section of all cooling channels. From physical tables [46], [47], [48] the following data can be found for water Dynamic viscosity : = 1.4 10 3 (kg/m s ) Heat conduction : = 0.57 (W/m K ) Water at 7C Specific heat : C p = 4194 (J/kg K ) Density : (kg/m3 ) d = 999.901 The Reynolds number can then be calculated to Re43000, which indicates that the fluid flow through the cooling channels are entirely turbulent. When the flow is turbulent, there is an engineering formula expressing the relation between an other dimensionless quantity called Nusselts number and Reynolds number [48]
Nusselts number : Nu =
c
d cc
Cp 0.023
0.4
Re 0.8
(5.9)
From this expression it is possible to write the convection heat transfer coefficient, c, as
81
Chapter 5
( 0.023
d) d 0.2 cc
0.8
Cp
0.4
0.6
(5.10)
Using this equation will give a heat transfer coefficient c33800 W/m2K which is a very high value. One assumption made in equation (5.9) is that L/dcc>200 where L is the length of one cooling channel. For the test cell L=170 mm and dcc=5 mm but when also considering the length of rubber tubing connecting the test cell this condition will be fulfilled. Experience shows that after a while there will be some kind of coating on the inside walls of the cooling channels due to oxidation. These coatings are difficult to characterise since their thickness and thermal properties are most often not known. To compensate for these effect the total heat transfer coefficient can be expressed as [48]
1
tot
1
c
1
coat
(5.11)
where coat is an experimentally determined value which is dependent on the type of cooling liquid used and the average speed of this liquid. The effect from the coating can be set to coat5600 W/m2K for heating plate temperatures Tsurface115C and coat2800 W/m2K for heating plate temperatures 115C<Tsurface205C [48]. This gives a total heat transfer coefficient for the cooling channels in this test cell of tot4800 W/m2K, Tsurface115C, and tot2600 W/m2K, 115C<Tsurface205C. The conclusion from these values is that the effect from the coating on the walls inside the cooling channels is very strong. Using equation (5.3) and that the wall area of each cooling channel is 0.0027 m2 the total cooling effect of each heating plate can be calculated, see Figure 5.5.
10000 Calculated cooling power [W]
5 cooling channels 1 cooling channel
8000
6000
4000
2000 Minimum 0 0 50 100 150 200
Figure 5.5
Calculated cooling power for one heating plate as a function of heating plate temperature. Minimum requirements for the test cell is 420 W cooling power.
To meet the minimum factory requirement for cooling, 10C/min, the cooling channels have to remove at least 420 W. In Figure 5.5 it is seen that even one cooling channel is sufficient for the higher temperatures but to be on the safe side it was chosen to use five cooling channels for each heating plate. The cooling system could be optimised more working with parameters like number of cooling channels and the temperature of the cooling liquid. However, this cooling system is sufficient for this application.
82
Chapter 5
There is no consideration taken here for heat conduction effects due to condensation or evaporation of the cooling liquid involved. Such phenomena will of course occur for heating plate temperatures above 100C.
5.3
Controlling the temperature
An Eurotherm 2408P4 [23], high stability PID temperature controller, regulates the temperature path of the test cell, see Figure 5.6. It was factory adjusted to have two inputs measuring temperature and two outputs regulating heating and cooling. One thermocouple was drilled into each heating element. The readings from the two thermocouples, upper and lower heating plates, were connected in such a way that the Eurotherm was measuring their average value. This value was then compared with the setpoint value programmed in the Eurotherm and the difference generated via the PID-parameters heating or cooling. Heating took place by pulsing power (Pmax670W) to the heating plates via a solid state relay connected to a 2-pin logic output on the Eurotherm. Cooling took place by opening and closing a magnetic valve with 3 bar cold tap water (7C). The magnetic valve was switched by a solid state relay that was controlled by a 2-pin relay output on the Eurotherm.
Cold tap water Heating element
Thermocouple Magnetic valve Solid state relays Input Eurotherm 2408 P4 Output AC 230V RS232
Figure 5.6
The set-up for controlling temperature paths in the test cell. The temperature of each heating element is measured by an Eurotherm (PID regulator) which then regulates heating or cooling.
Some of the PID-parameters, tuning parameters, in the Eurotherm can be set either by oneshot, adaptive or manual tuning. With tuning means that the tuning parameters (characteristics of the controller) are matched to the characteristics of the process controlled (response of the test cell and test sample). Good tuning means that there is straight-line control of the temperature without fluctuation and no overshoot or undershoot of the temperature setpoint. With one-shot tuning is meant turning the two outputs, heating and cooling, on and off inducing an oscillation in the measured temperature. From the amplitude and period of the oscillation, the tuning parameters are calculated. One-shot tuning can be performed at any time and is preferable done before the temperature path of the test cell is changed a lot. Adaptive tuning is done with a background algorithm that continuously during process updates the tuning parameters. Adaptive tuning should be used with care but can be 83
Chapter 5
interesting when processing test samples like the resin-rich insulation since it changes its thermal properties during curing. If for any reason the automatic tuning does not work it is possible to tune the Eurotherm manually. There are a number of standard methods to chose for manual tuning. It is important to tune the Eurotherm and all three methods have been used but for different test samples and for different temperature paths. The Eurotherm could be manually controlled or pre-programmed from a computer via a serial RS232 connection. All programming was done with software specially written for Eurotherms called IPS that was running under the DOS operating system. This Eurotherm could store four programs with each 16 segments. There are five different segments, ramp, dwell, call, step and end to chose from when programming the temperature paths, see Figure 5.7. The ramp segment will ramp the temperature up or down at the set rate whereas the step segment will instantly change the temperature up or down. The dwell segment will keep the temperature constant for a given period compared to the end segment that will either end the program or repeat the program again. There is also a segment, call, to call other subroutines (programs) that are executed and then return control to the following segment in the main program. Subroutine Temperature Ramp Dwell Call Step End t0 t1 t2 t3 t4 Time Figure 5.7 Examples of the different segments available when setpoint programming the Eurotherm 2408P4. It was possible to store four programs with 16 segments each.
Before programming a temperature path the tuning parameters (PID-parameters) should be tuned to this profile. For this test cell one-shot tuning was made in the temperature region 110C to 170C. To get a feeling for how the test cell was working, two steps, one heating followed by one cooling, was programmed and the temperature of the top and bottom heating plate was measured. In Figure 5.8 the result from this two-step program can be seen for the low, (A), and the high, (B), temperature region. The maximum heating rate is found to be in the low region 9.7C/min compared to 8.3C/min in the high region. This was expected since the heat losses will increase with increasing temperature, see Figure 5.4. The cooling part is more difficult and some oscillations are found in the measurements. For the low temperature region, (A), the measured values first drop rapidly and then slowly reach the final value without any oscillations. In the high temperature region, (B), there is an undershoot of about 10C followed by an oscillation, 3C, for about 10 min. The tuning parameters were the same for both measurements.
84

90 80 Temperature [C] 70 60 +9.7 C/min 50 40 30 20 10 0 20 40 60 80 100 80 60 (A)
Program Meas. Top Meas. Bottom
Chapter 5
180 (B) 160 Temperature [C] 140 +8.3 C/min 120 100
20
Time [min]
40 60 Time [min]
80
100
Figure 5.8
The temperature response of the test cell, top and bottom heating plate, to a programmed step in the low temperature region, (A), and the high temperature region, (B).
Not only the maximum heating and cooling rates are interesting it is also important to know how the test cell is responding to temperature ramps of different slopes. In Figure 5.9 (A) it is seen how the test cell responds to heating with rates of 0.5, 2 and 8C/min. Heating is done beginning from 20C up to 180C starting both with rates of 0.5 and 8C/min. All heating ramps are following their setpoint values well but it seems that low heating rates at low temperatures are the most difficult for the test cell. In Figure 5.9 (B), it is seen how the test cell responds to cooling from 180C down to 20C with the same rates as for heating. As for heating, cooling at these different rates work well except for some oscillations also here at low cooling rates and at low temperatures.
200 (A) 2 C/min Temperature [C] Temperature [C] 150 8 C/min 100 8 C/min 50 2 C/min
0.5 C/min
200 -0.5 C/min (B) 150 -8 C/min 100 -8 C/min 50 -2 C/min 0 -0.5 C/min 0 20 40 Time [min] 60 80 -2 C/min
0 0 20
0.5 C/min 40 Time [min]
60
80
Figure 5.9
The test cell temperature response, top and bottom heating plate, to programmed ramps, 0.5, 2 and 8C/min. (A) response to heating and (B) response to cooling.
However, both heating and cooling according to the ramps studied here work well. The test cell will also meet the minimum factory requirements of a heating rate of 5C/min and a cooling rate of 10C/min. It is concluded that it is better to use ramps when programming 85
Chapter 5
temperature changes, also very rapid ones, since this will reduce undershooting and oscillations giving more stable temperature control.
5.4
Controlling the pressure
When producing resin-rich insulation systems pressure is an important parameter. With pressure it is possible to change the dimension of the insulation and thereby make the stator bars fit into the slots in the stator core. Especially in the beginning of the process when the resin becomes liquid it is important to press in order to equally distribute the resin and also with the resin remove moisture if existing. For this test cell, it was determined to use only static pressure. The intention was to study if pressure had any influence on the dielectric response and if so how much. In the factory process pressures of about 1 MPa are involved which means that for our test cell, 170mm170mm, a force of approximately 30 kN is needed. The easiest way to generate this force in a consistent way is to build a frame around the test cell that is bolted together with several bolts, see Figure 5.10.
F1 Ftot=F1+F2 Heating plate Sample Heating plate F2
Figure 5.10
Pressing the test sample with a static force, Ftot, generated from up to 12 bolts. Each bolt had a number of conical washers building up a spring.
Each bolt should have a number of conical washers building up a spring allowing each bolt to be tightened equally hard and giving the same pressing force. Depending on how the conical washers are put together on each bolt the resulting spring will have different force-length characteristics, see Figure 5.11 (A). (A) Force 3F 2F F k2 (I) k1 s Figure 5.11 2s 3s k3 4s Length (III) t (II) (B) F = kn s Di h0 De l0
(A) The characteristics from a spring built up by conical washers in different ways. (B) The dimensions from one washer, height l0=2.45 mm, dynamic height h0=0.7 mm, outer diameter De=31.5 mm, inner diameter Di=16.3 mm and thickness t=1.75 mm. 86
Chapter 5
Case (I) represents the behaviour of one linear washer, case (II) is the result when putting three washers on top of each other in the same direction and case (III) is the result from putting four washers together with every second in the opposite direction. The idea when building up the spring was that a change in spring length should result in a small change in force, see case (III). This was important since during processing of resin-rich insulation there is shrinkage due to curing. Each spring used with the test cell was built up from 38 conical washers in the way shown in case (III). The force-length characteristics for each washer used in the spring can be seen in Figure 5.12. Each washer had, unpressed, a dynamic height h0=0.7 mm, total height l0=2.45 mm, outer diameter De=31.5 mm, inner diameter Di=16.3 mm, thickness t=1.75 mm and the active (linear) region was between 10% to 75% of h0. The washers were chosen to have a linear characteristic because that gave simple calculations. If the dimensions of the test sample change a lot during the measurement, it could be useful to use washers with a nonlinear characteristic. However, the assumption was made that the shrinkage due to curing of the resin-rich insulation was to be small and therefore a linear characteristic was easier to handle.
5000
Measured Linear curvefit
Max
4000 Pressing force, F [N] 3000 Min 2000
l0
F=F(s) s=0 s=smin Active region s=smax
1000 0
s=nl0
0 0.2 0.4 0.6 0.8 1
Normalised spring length, s/h 0
Figure 5.12
Characteristics, force-length, of each washer used in the test cell. Each spring used n=38 washers and their active (linear) region was between 10% to 75% of their total dynamic height nh0.
For example when the test sample, A0.122 m2, should be pressed with 1 MPa the force F14.4 kN was needed. This force was then divided in between four springs, F1=F2=F3=F43600 N. Then from the graph in Figure 5.12 a force of 3.6 kN will give s/h00.69 which will give a total compression of n=38 washers of stot18.4 mm. This means that the four springs should be screwed to a height Ltot=nl0-stot74.7 mm. Assume now that you have a test sample that is around 5 mm thick and that there is a shrinkage of about -5% during the measurement. This will give a new total height for each spring of Ltot74.7+0.25 mm which will give s/h00.68. The new force can then be read from the graph to be F1=F2=F3=F43540 N. So for a 5 mm thick test sample shrinking -5% the new pressure will be 0.98 MPa which is a -2% change.
87
Chapter 5
5.5
Measurement circuit
The first step after that the test cell was designed and assembled was to electrically connect everything for the dielectric response measurements. The setup is very simple [60], see Figure 5.13. Noise and drift can be introduced into this setup in four ways: (1) by the signal source (voltage source), (2) by the connections to the measuring instruments (electrometer for both voltage and current), (3) in the measuring instruments, and (4) by external disturbances from electromagnetic fields. The effects responsible for the most significant levels of noise and drift in this setup are from the connections, (2), and by external disturbances, (4).
Screen (closed loop)
+ U Loop area small
Test sample (guarded)
I Ground (only one)
Figure 5.13
Basic setup for the dielectric response measurements. All couplings were made with as short as possible screened low noise cables that were twisted. Ground loops were avoided and the test sample was screened and guarded.
All connections were made with screened low noise coaxial cables that were twisted in order to reduce noise due to magnetic fields. Twisting the cables is also preferable when they are carrying large currents to prevent generating magnetic fields that influence other components in the measurement circuit. Moving cables in weak static magnetic field (even the earths magnetic field of 62 T) can also generate noise, so cables must be kept short and rigidly fixed down. Noise and error voltages also occur from so called ground loops. A ground loop is generated when a measurement circuit has several points that are grounded. These ground loops can carry large currents generating unwanted voltage drops. To avoid this it is important to isolate each measurement instrument and only ground the test circuit at one point. Thermoelectric voltages are also common errors in voltage measurements when different parts of the measurement circuit are at different temperatures. This is of course something that is inevitable in this setup, but the effect will be reduced by using the same material for all conductors. Electrochemical effects can arise in the test sample during curing and also when measuring at different temperatures. This problem is difficult to solve and in most cases something one has to accept. In the measurement circuit outside the test cell electrochemical effects can occur in for example dirty contacts which can be avoided by cleaning with isopropanol. In Figure 5.14 the actual setup for dielectric response measurements in the frequency domain is shown. For frequencies 102<f<106 Hz a Hewlett Packard 4284A precision LCR-meter was used measuring capacitance and loss at 9 V and for 10-4<f<102 a dielectric response analyser was used measuring at voltage levels up to 200 V [28], [37], [98]. The measurement connection to the test cell is a 4-wire terminal connection that is used in order to reduce the influence of the test cables especially for higher frequencies. The heating plates in the test cell are guarded and the test cell itself is insulated from ground potential to avoid ground loops. 88

HP 4284A LCR meter
H Cur H Pot L Cur L Pot
Chapter 5
< 1 M Hz > 100 Hz
4-wire terminal
Test cell Heating 230V Thermocouples
Dielectric response analyser
Hi Low
< 100 Hz f > 0.1 mHz
Insulated support Guard
230V
Cooling
Figure 5.14
Measurement setup for dielectric response measurements in the frequency domain during curing of resin-rich insulation.
Other factors that will contribute to the noise and drift level are the thermocouples, heating and cooling. Thermocouples could act as pick-up antennas for high frequency noise but numerous experiments tend to indicate that this is not the case. The two thermocouples that measure the temperature paths of the upper and lower heating plates are electrically insulated with a thin PET-film from the heating plates to avoid ground loops. Heating is done via switching of a solid state relay, pulsing the power to the heating plates. This switching generates transients that can be seen in the current through the test sample, see Figure 5.15. The magnitude of the current transients is dependent on the capacitance between the heating elements and the electrodes and on the capacitance of the test sample. It is seen in the figures that the transients are about 2.5 times higher in magnitude than the 100 Hz background noise and that they are in the 40 kHz frequency range. These transients seem to be so small that they do not affect the measurements in a critical way.
Figure 5.15
Transient from the solid state relay switching during heating of the test cell. Measured at T=100C with a test sample of C830 pF with a current amplifier, Keithley 428, at gain 106.
Transients are also seen in the current through the test sample due to switching of the magnetic valve controlling the cooling water see Figure 5.16. These transients are much larger in magnitude than the one from the solid state relay and they come in pairs of two of which the first one is the most severe one. It is seen in the figures that the first transients are about 35 times higher in magnitude than the 100 Hz background noise and that they are in the 30 kHz frequency range. The magnetic valve is screened so the transients are probably coupled via the water in the cooling channels into the heating plate. These transients cause problems for measurements at low frequencies, f<0.1 Hz, around room temperature, T<30C, since they cause the current measuring electrometer to go into overload. A way around this problem is to 89
Chapter 5
integrate the current. The cooling system has also the disadvantage that it introduces a ground loop via the waterways when cooling.
Figure 5.16
Transients from switching of the magnetic valve controlling the cooling water for the test cell. Measured at T=20C with a test sample of C750 pF with a current amplifier, Keithley 428, at gain 106.
What is clearly seen in both Figure 5.15 and Figure 5.16 is the 100 Hz noise level picked up by the measurement circuit. Looking carefully at the 100 Hz noise level at the two different temperatures, T=20C and T=100C, it is seen that the noise level is reduced for the higher temperature. Before the test cell was used for pressing the resin-rich samples studied in this thesis benchmark measurements were performed on a 0.1 mm thin PTFE film at T=22C, see Figure 5.17.
(A) T=22 C
HP 4284A Diel. resp. analyser: K6517 Diel. resp. analyser: K428 Literature: von Hippel
Real part of the complex permitivitty
10
-2
2.15 (B) T=22 C 2.13

HP 4284A Diel. resp. analyser: K6517 Diel. resp. analyser: K428 Literature: von Hippel
10 tan()
-3
2.11
10
-4
2.09
2.07
10-5 -3 10
10
-1
10
2.05 10-3
10-1
101
103
105
Frequency [Hz]
Figure 5.17
Measurements of tan and real part of the complex relative permittivity for PTFE at T=22C compared with values found in the literature [32],[40]. For frequencies f>100 Hz the HP 4284A was used and for frequencies f<100 Hz the dielectric response analyser was used. The dielectric response analyser was used both with capacitive (K6517) and resistive (K428) feedback in order to measure the current.
Measured values of tan and real part of the complex relative permittivity were compared to values found in the literature [32],[40]. What can be seen is that there is a wide spread in the 90
Chapter 5
measured data but they are all in the right order of magnitude and have the right overall behaviour. The calculated real part of the complex relative permittivity will also be very sensitive to the geometric capacitance. It is important to remember that this benchmark was made on a very low loss material that is known to be difficult to measure. The actual test samples will have a loss level that even for the fully cured system will be several decades higher in magnitude. The resin-rich mica tape samples were all first mounted in the test cell with the right pressure and then electrically connected to the measurement circuit. Each measurement starts with programming the temperature path into the Eurotherm. Then from time to time a calibration measurement against a polypropylene capacitor was performed to check the setup. Depending on the temperature path and on what polarisation phenomena was being studied the frequency region for the measurement was chosen. When measuring on resin-rich insulation the basic idea was to follow the cross-linking by shifting the frequency window towards lower frequencies as curing continues. An example of this is seen in Figure 5.18 where the measurement during Part (I) uses the HP 4284A LCR-meter and during Part (II) uses the dielectric response analyser. Going from Part (I) to Part (II) the measurement frequencies are shifted three decades towards lower frequencies. It is also seen that the features of the test sample are changing in the same way, the measurement follows the changes in the test sample.
Temp [C] Part (I) 120 Part (II)
20 3.1 3.3 log(Time) [s]
Figure 5.18
Example of a tan measurement that were split up measuring first at higher frequencies, with the HP 4284A, and then at lower frequencies, with the dielectric response analyser. By doing this, it is possible to follow the crosslinking in the test sample.
When measuring at constant temperatures the test sample was always short-circuited when changing to a new temperature in order to remove all charges building up due to thermal gradients.
91
Measurements on resin-rich insulation
Chapter 6
6.1
Introduction
In this chapter, the dielectric response in frequency domain will be measured on two different types of resin-rich mica tapes. The first tape has woven glass and the second polyester film (PET) as carrier material. This type of insulation systemis used as electrical insulation in highvoltage rotating machines. The main idea was to use non-destructive dielectric response measurements to study the manufacturing stage of these resin-rich mica tapes. Low voltage measurements, 5 to 200 V peak, were made in the frequency range 10 mHz to 1 MHz using the test cell described in chapter 5. Two questions were raised from the beginning. What would the change in dielectric response look like during curing and was it possible to read out a specific degree of curing from a dielectric response measurement at constant temperature. Simple parallel-plate samples were pressed and cured under varying conditions in order to investigate this. The dielectric response was measured before, during, and after the processing cycle. Experimental results were compared to a full-scale factory measurement. A simple idea was stated of how the rate of cross-linking in the epoxy-resin would affect the quality of the insulation in terms of built in tension and adhesion between layers and towards the conductor. A high rate of curing would quickly freeze the system leaving out possible cross-links whereas a slower rate would ensure that all cross-links would find each other. To investigate this several different temperature paths were used including processing under constant temperature. A similar temperature path as recommended by the tape manufacturer together with one fast to get a high rate of cross-linking and one slow to get a low and homogenous rate of cross-linking were chosen. In order to try verifying that the slow temperature path compared to the fast gave better mechanical properties a simple mechanical test of the adhesion between the copper conductor and the insulation were performed. The chemical reaction during curing of the resin-rich mica tape was also studied with differential scanning calorimetry (DSC). To determine the degree of curing for a sample, a DSC measurement for this sample was related to a DSC measurement on an uncured sample.
6.2
Preparing test samples for a dielectric response measurement
Since resin-rich insulation contains epoxy, it is important to store this type of insulation in a cold place to avoid pre-curing. The storage life of the insulation is about 7-12 months in a 93
Chapter 6
refrigerator (+5C) and 3-6 months at room temperature (+22C). When storing the insulation in a freezer, <0C, it is important to avoid humid air crystallising into ice on the insulation surface. This ice will later melt when preparing the test sample and the moisture might influence the measurements. The test samples were built up from sheets, approximately 120120 mm2, of the resin-rich insulation stacked on top of each other. Depending on the purpose of the measurement between 4-30 sheets were used. The different types of resin-rich insulation studied here had often a carrier material like woven glass or a PE-film on one side. Therefore, when stacking the sheets they were sorted in such a way that the carrier materials always were put upwards. The electrodes were cut out from 0.2 mm thick brass sheet with a diameter of dV=110 mm for the voltage electrode and dC=90 mm for the current electrode, see Figure 6.1.
Top/Bottom view Voltage electrode 4 mm Current electrode Guard
120 mm
dV=110 mm
dC=90 mm
PTFE tape 120 mm
Figure 6.1
The top and bottom view of a prepared test sample, 120120 mm2. Voltage and current electrodes made of brass with diameters of dV=110 mm and dC=90 mm. A guard was made with copper tape with a guard distance of about 2 mm.
Each electrode was cut almost circular except for a 4 mm wide tangential string that was used as a dielectric response measurement connection. Before the electrodes were added to the stack of resin-rich sheets they were polished and cleaned with isopropanol. On the side of the current electrode, a 2mm wide guard was applied using copper tape. The guard is to protect the measurements against leakage current [99], [104]. However, the guard is only needed for the fully cured system where the losses are low. Everything was held together with a few pieces of PTFE-tape in each corner of the test sample. The side view of the test sample is shown in Figure 6.2. Up to 20 thermocouples can be inserted in the test sample to measure the temperature gradient throughout the measurement.
Side view U Electrodes I Test sample PET film Thermocouples PET film
Figure 6.2
Side view of a prepared test sample where the temperature path is measured with 0.5 mm thick thermocouples slightly shifted not to be pressed together. PET films are used to protect the test cell against epoxy. 94
Chapter 6
Each thermocouple is of type K and is 0.5 mm thick with an insulation that can withstand up to 260C. In order not to press the thermocouples together, they are slightly shifted both sideways and in the vertical direction. To electrically insulate the measurement electrodes from the heating plates a 0.05 mm thick PET-film was used. The PET-film that had a melting point of 265C also served as protection for the heating plates against curing epoxy. The whole procedure to assemble a test sample takes around 30-40 minutes followed by 5-10 minutes to fix the test sample in the test cell and then the measurement can start.
6.3
Preparing test samples for a differential scanning calorimetry (DSC) measurement
With differential scanning calorimetry (DSC), it is possible to study phase transitions or different chemical reactions in a polymeric material when subjected to heating, cooling or a constant temperature. The basic idea is to measure the heat flow difference between an inert reference and a test sample during a pre-programmed temperature path. From the heat flow to the test sample, as a function of applied temperature path it is possible to identify specific features like changes in heat capacity, glass transitions, crystallisation (exothermal reaction), melting (endothermal reaction) and different chemical reactions [26]. An example of a chemical reaction is the curing of an epoxy resin where the network formation is an exothermal process, which can be studied by means of DSC. It is generally assumed that the exothermal heat released is proportional to the number of bonds that have reacted. The DSC measurements were made with two different instruments, one Perkin-Elmer DSC7 and one Mettler-Toledo 820. The two instruments were calibrated according to instructions from the manufacturer. The accuracy of the measured relative heat flow for the two instruments is on the order of a W. The test samples were cut from the resin-rich mica tape taken directly from the refrigerator with a pair of scissors, which was cleaned with isopropanol. Two to three pieces were put in each aluminium sample holder, 30 l, and pressed together to ensure a good thermal contact before the lid was put on. Measurements were made both with and without a small hole in the lid. When a hole in the lid was used 80 ml/min of nitrogen gas was used to ventilate the sample. Each sample had a total weight without sample holder of about 6-10 mg and was subjected to a heating rate of 9C/min in the temperature range 20C to 300C. To determine the degree of curing for the resin part in the resin-rich mica tape the total exothermal energy, Qtot, in the temperature range 20C to 300C (ramp 9C/min) is first measured for a sample of the uncured tape. A new sample is then cured under a preprogrammed temperature path, T=T(t). The remaining exothermal energy, QT, is then measured in the same way as for the uncured tape. The degree of curing, , is then defined as [70]
= 1 QT Q tot
(6.1)
The DSC measurements made here on the two types of resin-rich mica tapes will of course reflect the total heat flow to the whole composite. This means that the temperature
95
Chapter 6
dependences and the possible chemical reactions from other materials than the resin in the composite will be measured.
6.4
Woven glass as carrier material
The resin-rich mica tape used here consists of a flexible mica paper foil, which is thoroughly impregnated with an electrical grade modified epoxy to which a woven glass carrier material is bonded to provide sufficient mechanical strength [87]. The mica paper foil, which is of muscovite type, will provide a high electrical resistance to corona. This tape is designed to be used in class F rotating machines with a operating temperature of 155C but according to some manufacturer it can be used up to 180C. The tape is used both as conductor and turn insulation and as main wall insulation. It is wrapped either by machine or by hand with a tension of 40-60 N and the individual layers are applied half-lapped. There is a compression ratio of about 25-30% to produce the final dimensions, giving recommended operating electrical field strength of max 3 kV/mm. The chosen insulation thickness depends not only on the operating voltage but also on the mechanical stresses. When pressing the insulation, at 2.0-3.0 N/mm2, it is recommended that the press be preheated to around 100C, before loading the coil into the press and then to wait until the whole coil section in the press reaches at least 100C. The temperature is then raised to the final curing temperature, which according to the manufacturer of the tape should not exceed 180C. After pressing the coils, they are put into the stator core and a post-curing will take place at around 140C for 12-16 hours. In Figure 6.3, two microscopic pictures are shown from the resin-rich mica tape discussed in this section. There is a clear repeating rectangular pattern, which consists of the woven glass threads. The mica flake surface areas in the foil are parallel with the tape surface giving the greatest dielectric strength, can also be seen in the right picture. It is possible that during the curing process there will be epoxy, moving through the woven glass carrier.
Figure 6.3
Microscopic pictures of the resin-rich mica tape with woven glass as carrier material. There is a clear structure in the tape, a rectangular pattern from the woven glass carrier. The surfaces of the mica flakes that are all parallel with the surface of the tape.
The uncured tape has a thickness of 0.18 mm and a total weight per unit area of 26526 g/m2. It is composed of 1207 g/m2 mica, 323 g/m2 glass fabric and 11320 g/m2 resin. After curing and pressing, the composite insulation has a density of 1.8-2.0 g/cm3, dielectric 96
Chapter 6
strength at 23C of >35 kV/mm, dissipation factor at 23C of 0.8-1.2% (10kV, 50 Hz) and a thermal conductivity at 23C of 0.28 W/mK.
6.4.1
The uncured system
The dielectric response of the uncured resin-rich mica tape was measured in the frequency range 10 mHz to 1 MHz and in the temperature range -50C to 50C under a constant pressure of 1.2 MPa. The measurement series started at room temperature, then was cooled down in steps to -50C, then heated back up to room temperature, repeating that measurement, and then increased up to 50C. The maximum temperature was chosen under the assumption that the test sample would still be uncured after the measurement. In Figure 6.4, capacitance and tan are plotted as surfaces giving a good overview of their frequency and temperature behaviour. The uncured tape shows a strong dispersion in both frequency and temperature.
Figure 6.4
Measured capacitance and tan for the uncured resin-rich mica tape with woven glass as carrier material as function of frequency and temperature.
In Figure 6.5, capacitance and tan are plotted as function of frequency for different temperatures. Strong temperature dependence in capacitance and tan is linked to the change in viscosity of the uncured tape. There is also a possible dipolar loss peak in tan which becomes visible in the measurement window for temperature above 0C. It is difficult to say if the measured behaviour in capacitance and tan is due to the different materials in the composite or due to their geometrical configuration, see Figure 6.3. There is a strong increase in loss for lower frequencies at the higher temperatures and there is even a tendency to electrode polarisation for the highest temperature [2], [71].
97
Chapter 6
10
-7 50 C 40 C 30 C 20 C 10 C 0 C -9.8 C -19.5 C -29.1 C -37.4 C -47.9 C

10
2 50 C 40 C 30 C 20 C 10 C 0 C -9.8 C -19.5 C -29.1 C -37.4 C -47.9 C
101 100
-1
10 C [F]
-8
tan() 10
-9 1 2 3 4 5 6
10
10 10-10 -2 -1 0 10 10 10
-2
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.5
The measured capacitance and tan of the uncured resin-rich mica tape. The uncured tape is strongly temperature activated and for temperatures higher than 0C there is a loss peak entering the measurement window.
6.4.2
The fully cured system
The test sample was first cured in the test cell starting at 20C with a rapid heating up to 80C and then increasing the temperature with 0.5C/min up to 160C keeping that temperature for 7 hours. This temperature path was chosen from the specifications given by the manufacturer also including the post-cure stage. The dielectric response of the fully cured resin-rich mica tape was measured in the frequency range 10 mHz to 1 MHz and in the temperature range 50C to 170C under a constant pressure of 1.2 MPa. In Figure 6.6, the capacitance and tan are plotted as surfaces giving a good overview of their frequency and temperature behaviour. The fully cured tape shows a weak dispersion in both frequency and temperature for temperatures lower than 70C. There is also a minimum in tan passing the measurement window for the temperature range -50C to 20C.
Figure 6.6
Measured capacitance and tan for the fully cured resin-rich mica tape with woven glass as carrier material as a function of frequency and temperature.
In Figure 6.7, capacitance and tan are plotted as function of frequency for different temperatures. During the curing of the resin in the tape, the network formation will increase the glass transition temperature, which is the same as a decrease in the mobility of the 98
Chapter 6
polymer chains. This is also seen in the dielectric response, which is low for the lower temperatures. When increasing the temperature, which is the same as going downwards in frequency, a loss peak is shifted into the measurement window representing rotational polarisation losses of large parts of the polymer network.
10
-8 170 C 150 C 130 C 110 C 90 C 70 C 50 C 20 C 10 C -9.7 C -29.2 C -47.6 C
10
0 170 C 150 C 130 C 110 C 90 C 70 C 50 C 20 C 10 C -9.7 C -29.2 C -47.6 C
10-1 tan() 10-2 10-3 -2 -1 0 10 10 10
C [F]
10-9
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.7
The measured capacitance and tan of the fully cured resin-rich mica tape. There is a loss peek in tan entering the frequency window for temperature higher than 100C.
However, it is also in this case difficult to say if the measured behaviour in capacitance and tan is due to the different materials in the composite or due to their geometrical configuration.
6.4.3
Different degrees of curing
In order to investigate the behaviour of the dielectric response for different degrees of curing four resin-rich mica tape samples were cured under different temperature paths but with constant pressure, 1.2 MPa. After each profile, two DSC measurements were made for each sample to determine the degree of curing. The two samples for the DSC measurements were taken from two different positions on the main sample. The results from the DSC measurement, total remaining exothermal energy QT, were then compared to the total exothermal energy, Qtot, in an uncured tape. The first step before curing each of the four samples was to measure DSC on the uncured resin-rich mica tape. In Figure 6.8, the results are shown from six DSC measurements taken from two different production dates and from three different positions on the tape. The reproducibility is good especially since this is a composite with a spread in distribution of the different materials in the composite. The total exothermal energy per gram of the resin-rich mica tape was estimated from a scaled baseline to be Qtot=14816 J/g. The baseline was scaled, meaning that it was estimated from the heat capacity for the uncured and cured state and in between scaled with the degree of curing.
99
Chapter 6
0.6
1 2 3 4 5

0.6
M-T, m=8.73 mg Scaled baseline
Heat flow per gram [W/g]
0.4
0.4
0.2
0.2
0
1) M-T, m=2.05 mg 2) P-E, m=8.85 mg 3) P-E, m=8.53 mg 4) M-T, m=8.73 mg 5) P-E, m=8.60 mg 6) M-T, m=7.24 mg
-0.2
-0.2
Qtot = 148+/-16 [J/g]
-0.4 50
100
150
200
250
300
-0.4 50 100 150 200 250 300 Temperature [C]
Temperature [C]
Figure 6.8
DSC measurements on uncured resin-rich mica tape insulation. The total exothermal energy per gram sample was estimated to be Qtot=14816 J/g.
Four samples were then cured according to the temperature paths shown in Figure 6.9 (A). The temperature paths start at 20C, heating rapidly up to 80C and then changing heating rate to 0.5C/min up to Tmax=116, 126, 136, 151C respectively. Cooling was then done at 0.5C/min down to 80C and then maximum cooling down to 10C. These profiles were chosen more or less on a trial and error basis. In Figure 6.9 (B), the DSC measurements are plotted for the uncured tape and for the four samples cured according to the temperature paths described above. There is a clear shift and reduction in exothermal energy with increasing degree of curing. The degree of curing was then calculated for Tmax=116, 126, 136, 151C to be 0.4, 0.6, 0.8, 0.9.
160 140 Temperature [C] 120 100 80 60 40 20 0 0 50 100 150 200 250 300 350 Time [min] -0.4 50 100 150 200 (2) (3) (4) (5) 151 C 136 C 126 C 116 C 0.6
0.4
0.2
3 4
5 2 1
1) Uncured, =0 2) Calc. = 0.38 3) Calc. = 0.58 4) Calc. = 0.75 5) Calc. = 0.89
-0.2
250
300
Temperature [C]
Figure 6.9
(A) The different measured temperature paths used to cure the resin-rich mica tape. (B) Measured DSC for the partially cured samples and from that the degrees of curing are calculated.
6.4.3.1
Degree of curing, 0.4
The capacitance and tan were measured for the sample that was cured to 0.4 in the frequency range 10mHz to 1 MHz changing the temperature from -50C to 50C. In Figure 100
Chapter 6
6.10, capacitance and tan are plotted as surfaces to give an overview of the dispersion in frequency and temperature. A similar ridge in tan as seen for the uncured tape, see Figure 6.4, is also seen here but shifted to lower frequencies.
Figure 6.10
Measured capacitance and tan as function of frequency and temperature for the partially cured resin-rich mica tape. The degree of curing was measured with DSC to be 0.4.
In Figure 6.11 capacitance and tan are plotted as function of frequency for different temperatures. The ridge in tan is seen in this measurement window for temperatures higher than 20C. These measurements above 20C have roughly a frequency shift of about three decades compared to the uncured tape, 0, see Figure 6.5. There seems to be two different activation energies in this system, one for temperatures above 0C and one below. For temperatures lower than -10C there is also a minimum appearing indicating a high frequency loss peak in the material in the sample.
10-8
50 C 40 C 30 C 20 C 10 C 0.2 C -10.0 C -20.0 C -29.0 C -39.2 C -48.1 C
100
50 C 40 C 30 C 20 C 10 C 0.2 C
-10.0 C -20.0 C -29.0 C -39.2 C -48.1 C
10-1 tan() 10-2
C [F]
10
-9
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 10-2 10-1 100
101
102
103
104
105
106
Frequency [Hz]
Frequency [Hz]
Figure 6.11
Measured capacitance and tan as for a resin-rich mica tape sample cured to 0.4. There seems to be two different activation energies in this system one above 0C and one below.
6.4.3.2
The capacitance and tan were measured for the sample that was cured to 0.6 in the frequency range 10mHz to 1 MHz changing the temperature from -50C to 50C. In Figure 101
Chapter 6
6.12, capacitance and tan are plotted as surfaces to give an overview of the dispersion in frequency and temperature. The ridge seen in tan for the uncured tape, see Figure 6.4, and for the tape cured to 0.4, see Figure 6.10, is not seen here. The ridge is probably shifted out of the measurement window towards lower frequencies.
Figure 6.12
Measured capacitance and tan as function of frequency and temperature for the partially cured resin-rich mica tape. The degree of curing was measured with DSC to be 0.6.
In Figure 6.13, capacitance and tan are plotted as function of frequency for different temperatures. In tan there seems to be two different activation energies with a transition around 15C. In the same way as the ridge seen for the uncured and 0.4 tape is shifted towards lower frequencies out of the measurement window a high frequency loss peak seems to be shifted in. This high frequency loss peak is seen for temperatures lower than 10C.
10
-8 47.7 C 37.3 C 30.1 C 20 C 9.8 C -0.1 C -10.1 C -19.4 C -29.4 C -39.1 C -57.0 C
100
47.7 C 37.3 C 30.1 C 20.0 C 9.8 C -0.1 C -10.1 C -19.4 C -29.4 C -39.1 C -57.0 C
10-1 tan() 10-2
C [F]
10-9
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 10 10-1 100
101
102
103
104
105
106
Frequency [Hz]
Frequency [Hz]
Figure 6.13
Measured capacitance and tan for a resin-rich mica tape sample cured to 0.6. There seems to be a transition between two different activation energies in this system around 15C.
6.4.3.3
The capacitance and tan were measured for the sample that was cured to 0.8 in the frequency range 10mHz to 1 MHz changing the temperature from -50C to 50C. In Figure 6.14, capacitance and tan are plotted as surfaces to give an overview of the dispersion in 102
Chapter 6
frequency and temperature. The capacitance and tan surfaces have similar shapes as for the sample that was cured to 0.6 except a bit lower levels for temperatures above T=20C.
Figure 6.14
Measured capacitance and tan as a function of frequency and temperature for the partially cured resin-rich mica tape. The degree of curing was measured with DSC to be 0.8.
In Figure 6.15, capacitance and tan are plotted as function of frequency for different temperatures. Looking at tan there seems to be only one activation energy in this temperature range and there is a slight shift in the minima towards lower frequencies compared to the sample cured to 0.6. There is almost no change in capacitance with frequency or with temperature.
10
-8 49.5 C 39.5 C 30.1 C 19.8 C 9.8 C -0.5 C -10.0 C -19.2 C -29.8 C -39.2 C -50.3 C
10
0 49.5 C 39.5 C 30.1 C 19.8 C 9.8 C -0.5 C -10.0 C -19.2 C -29.8 C -39.2 C -50.3 C
10-1 tan() 10-2
C [F]
10
-9
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 10-2 10-1 100
101
102
103
104
105
106
Frequency [Hz]
Frequency [Hz]
Figure 6.15
6.4.3.4
The capacitance and tan were measured for the sample that was cured to 0.9 in the frequency range 10mHz to 1 MHz changing the temperature from -50C to 50C. In Figure 6.16, capacitance and tan are plotted as surfaces to give an overview of the dispersion in frequency and temperature. The capacitance and tan surfaces have similar shapes as for the
103
Chapter 6
samples that were cured to 0.6 and 0.8 and the levels are similar to the sample cured to 0.8 especially for temperatures lower than T=30C.
Figure 6.16
In Figure 6.17, capacitance and tan are plotted as function of frequency for different temperatures. Looking at tan there seems to be only one activation energy in this temperature range and there is a slight shift in the minima towards lower frequencies and also a tendency to decrease in the high frequency peek seen for the lower temperatures compared to the sample cured to 0.6. There is, like for 0.8, almost no change in capacitance with frequency or with temperature.
10
-8 49.6 C 39.0 C 30.4 C 19.0 C 9.5 C 0.7 C -9.7 C -19.8 C -29.5 C -39.3 C -48.6 C
10
0 49.6 C 39.0 C 30.4 C 19.0 C 9.5 C 0.7 C -9.7 C -19.8 C -29.5 C -39.3 C -48.6 C
10-1 tan() 10-2
C [F]
10
-9
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.17
6.4.3.5
Loss, tan, plotted against degree of curing
To summarise the results presented above the loss tan is plotted against frequency and degree of curing, in Figure 6.18 for the temperature T=20C and in Figure 6.19 for the temperature T=-20C.
104

10
0 =1 =0.9 =0.8 =0.6 =0.4 =0
Chapter 6
10
0 f=0.13 Hz f=1.5 Hz f=55 Hz f=1.2 kHz f=12 kHz f=96 kHz
T=20C 10 tan()
-1
10-1 tan()
T=20C
10
-2
10-2
10-3 -1 10
10
10
10
10
10
10
10-3
0.2
0.4
0.6
0.8
Frequency [Hz]
Degree of curing
Figure 6.18
The change in tan as a function of (A) frequency and (B) degree of curing for the resin-rich mica tape insulation at T=20C.
In Figure 6.18, it can be seen that the largest dynamic in tan is found for the lower degrees of curing. Around 0.6, the values more or less stabilise, indicating that the final structure of the polymer chains are reached and that for higher degrees of curing just more cross-links are added. At lower temperature, T=-20C see Figure 6.19, tan show a different behaviour with a minimum for the lower frequencies around 0.6 and for the higher frequencies an increase in loss with increasing degree of curing. There seems to be a high frequency peak, which is shifted towards lower frequencies as degree of curing increases. At room temperature this shift would be seen at much higher frequencies.
10
-1
T = -20C
=1 =0.9 =0.8 =0.6 =0.4 =0
10
-1
T = -20C
f=0.1 Hz f=2.2 Hz f=47 Hz f=1.2 kHz f=12 kHz f=80 kHz
tan()
tan() 10
0
10-2
10-2
10-3 -1 10
10
10
10
10
10
10
-3
0.2
0.4
0.6
0.8
Frequency [Hz]
Degree of curing
Figure 6.19
The change in tan as a function of (A) frequency and (B) degree of curing for the resin-rich mica tape insulation at T=-20C.
6.4.4
Dielectric response along the surface of the resin-rich mica tape
The idea here was to try to measure the dielectric response along the surface of the resin-rich mica tape with woven glass. These measurements gave understanding for the conductive 105
Chapter 6
mechanisms along the surface of the tape, which were important for the network model presented in chapter 7.4. The tape is not symmetric, on one side is the woven glass and on the other side is the mixture of epoxy-resin and mica flakes. A test sample was built up from 12 layers of tape orientated with the woven glass downwards, the structure is shown in Figure 6.20 using copper tape as electrodes [52], [53]. The electrodes were separated around 1 mm from each other and extra care was taken not to overlap electrodes in-between layers. The sample was put in the test cell and all measurements were done under constant pressure of 1.2 MPa.
Top/Side view 50 mm 1 mm 1 mm
150 mm
120 mm
2.5 mm
PTFE tape Voltage Current
2+10 layers
Figure 6.20
Electrode configuration for measuring the surface dielectric response of the resin-rich mica tape with woven glass as tape carrier.
Since the sample was pressed together it can be difficult to distinguish between volume and surface effect, especially after curing. However, the effect from the position of the woven glass, tape carrier material, was expected to be seen and also possible movements of the epoxy resin without the restriction of the woven glass structure.
6.4.4.1
The uncured system
First measurements were made in the test cell on the uncured sample, 0, with a constant pressure of 1.2 MPa. Measurements of capacitance and tan in the frequency range 10 mHz to 1 MHz for temperatures ranging from 10C to 50C are shown in Figure 6.21. The measured capacitance is much smaller and the dispersion is much stronger than compared with the bulk measurement made with an electric field perpendicular to the woven glass, see Figure 6.5. The measured tan shows similar behaviour, especially for the higher temperatures, as in Figure 6.5 but the loss level is higher. Although, tan looks similar the strong dispersion in capacitance will also give a strong dispersion in the imaginary part of the complex capacitance see Figure 6.22. Looking at the features of the complex capacitances, remembering that the two test samples have different geometries, the measurement parallel to the woven glass shows low frequency dispersion (LFD) compared to the measurement perpendicular which shows a combination of LFD and a possible peak due to barrier effects.
106

10-7 10-8
-9 10 C 20 C 30 C 40 C 50 C
Chapter 6
102
10 C 20 C 30 C 40 C 50 C
10 tan()
10 C [F]
10
10-10 10-11 10-12 -2 -1 0 10 10 10 10-1
10 10 10 10 Frequency [Hz]
10
10
10-2 -2 10 10-1 100
101
102
103
104
105
106
Frequency [Hz]
Figure 6.21
Capacitance and tan measured along the surface of uncured resin-rich mica tape with woven glass as carrier material. Strong dispersion is seen in both capacitance and tan.
T = 30 C =0
C', tape carrier C", tape carrier C', // tape carrier C", // tape carrier
10 Complex capacitance [F] 10
-6
(A)
102
(B)
-7
T = 30 C =0
tan, tape carrier tan, // tape carrier
10-8 tan() 101 102 103 104 105 106 10

-9
101
100
10-10 10-11 10
-12
10
-1
10-13 -2 10 10-1 100
10
-2
10-2 10-1 100
101
102
103
104
105
106
Frequency [Hz]
Frequency [Hz]
Figure 6.22
(A) Measured complex capacitance for voltage applied both perpendicular () and parallel (//) with the woven glass carrier material in the resin-rich mica tape. (B) Measured tan for the perpendicular () and parallel (//) cases.
6.4.4.2
The test sample was then cured in the test cell starting at 20C with a rapid heating up to 80C and then increasing the temperature with 0.5C/min up to 160C keeping that temperature for 7 hours. This temperature path chosen for curing was the same as for the fully cured system studied in chapter 6.4.2. The dielectric response of the fully cured sample was measured in the frequency range 10 mHz to 1 MHz and in the temperature range 10C to 110C all under a constant pressure of 1.2 MPa. In Figure 6.23, measured capacitance and tan with the applied electric field parallel to the woven glass tape carrier in the fully cured resin-rich mica tape are plotted. The fully cured tape shows a weaker but still some dispersion in both capacitance and tan. In tan there is a clear dipolar peak seen in the measurement window shifting towards higher frequencies as 107
Chapter 6
temperature increases, followed by a low frequency dispersion part increasing also with temperature.
10
-9 110 C 90 C 70 C 50 C 30 C 20 C 10 C
10
2 110 C 90 C 70 C 50 C 30 C 20 C 10 C
10 tan()
10 C [F]
-10
10
10
-11
10-1
10-12 -2 -1 0 10 10 10
10
10
10
10
10
10
10-2 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.23
10
-8
Capacitance and tan measured along the surface of fully cured resin-rich mica tape with woven glass as carrier material.
100
C', tape carrier C'', tape carrier C', // tape carrier C'', // tape carrier
T = 30 C =1
T = 30 C =1
tan, tape carrier tan, // tape carrier
Complex capacitance [F]
10
-9
10-1 tan() 10-2

-12
10
-10
10-11
10
(A) 10
-13
10
-2
10
-1
10
10
10
(B) 10-3 -2 -1 0 10 10 10
10
10
Figure 6.24
(A) Measured complex capacitance for voltage applied both perpendicular () and parallel (//) with the woven glass carrier material in the resin-rich mica tape. (B) Measured tan for the perpendicular () and parallel (//) cases.
6.4.5
Curing under different temperature paths
The idea here was to measure capacitance and tan during the curing process of the resin-rich mica tape samples. The dielectric response was measured in the frequency range 10 mHz up to 1 MHz and under constant pressure. The temperature paths used were curing under a constant temperature, curing under a constant temperature ramp and curing under similar temperature paths as found in the factory process. The aim was to see if it was possible to separate the temperature effects from the degree of curing effects in the dielectric response.
108
Chapter 6
6.4.5.1
Curing under constant temperature
The samples were all pressed in the test cell, see chapter 5, with a constant pressure of 6 kPa. The measurements started at 20C and then at t=t1 rapidly heating the sample, 9C/min, to the desired temperature which was reached at t=t2. In Figure 6.25 to Figure 6.28 the capacitance and tan are measured for the constant temperatures T=120, 140, 160, 180C. The initial ramp between t=t1 and t=t2 will make the resin inside the tape more and more liquid increasing the capacitance and tan. It is during this period difficult to separate the effects of curing and temperature in the dielectric response. For T=120C in Figure 6.25 there is for times t>t2 a saddle point in tan passing through the measurement window from higher towards lower frequencies. The same type of saddle point is seen for the measurements done with fixed degree of curing, 00.5, with the same measurement window as used here. This indicates that as curing goes on this saddle point is shifted from high towards low frequencies with a rate that is dependent of temperature.
Figure 6.25
Measured capacitance and tan during curing of a resin-rich mica tape sample. From t=t2 the temperature was constant at T=120C.
In Figure 6.26, the saddle point is still visible but it is moving faster through the measurement window and in Figure 6.27 and Figure 6.28, it is no longer seen. For the higher temperatures, T=160C and T=180C, the saddle point or equivalent degree of curing indicated by this saddle point is reached already during heating, t1tt2.
Figure 6.26
109
Chapter 6
Figure 6.27
If the results in Figure 6.28, assuming fully cured tape at the end of the measurement, are compared to the results from the fully cured tape in Figure 6.7, there is a difference because no peak in tan is seen in Figure 6.28. This can be explained with the difference in applied static pressure that is around 200 times higher in Figure 6.7 than in Figure 6.28. For the T=120C measurement there are signs of electrode polarisation for the lower frequencies directly after changing the measurement window to lower frequencies.
Figure 6.28
In Figure 6.29, the capacitance and tan at f=50 Hz are plotted against curing time for the constant temperatures T=120, 140, 160, 180C. After 200 min of curing, samples were taken and by using the DSC technique the degree of curing were found to be 0.8, 0.9, 1, 1 respectively. For the T=120C measurement there is a peak around t=150 min which for the T=140C is shifted to around t=80 min showing the increase in reaction rate when increasing the temperature by 20C. This peak disappears in the initial temperature ramp for T=160C and T=180C measurements.
110

10-8
120C, = 0.8 140C, = 0.9 160C, = 1 180C, = 1
Chapter 6
t2, 120C 10
1 120C, =0.8 140C, =0.9 160C, =1 180C, =1
2, 180C
f = 50 Hz
f = 50 Hz
10
-9
tan() t
2, 120C
C [F]
10
10-10
t2, 180C 0 50 100 Time [min] 150 200
10-1 0 50 100 Time [min] 150 200
Figure 6.29
Measured capacitance and tan at f=50 Hz as a function of curing time, from t=t2, at constant temperatures T=120, 140, 160, 180C. After 200 min DSC measurements were made for the four samples and the degree of curing was found to be 0.8, 0.9, 1, 1 respectively.
6.4.5.2
Curing under a temperature ramp
The samples were all pressed in the test cell, see chapter 5, with a constant pressure of 6 kPa. The measurements started at 20C and then at t=t1 heating the sample with rates of 0.33, 1, 3C/min, up to 200C reached at t=t3. At t=t2 a temperature of 100C is reached and this point is used as a reference point when comparing the different measurements. In Figure 6.30, the capacitance and tan are measured for the 0.33C/min ramp. The capacitance and tan are in the beginning increasing for all frequencies with temperature but around t200 min, they both start to decrease although the temperature is still increasing. In tan there is a ridge passing through the measurement window when tan is decreasing. This ridge is probably a combination of effects due to the resin cross-linking, changing viscosity and the geometry in the tape from the woven glass carrier and the mica flakes. A minimum is then reached in tan after which the increase is only related to the increasing temperature because the sample is already fully cured.
Figure 6.30
Measured capacitance and tan for the resin-rich mica tape during a constant temperature ramp increasing from 20C with 0.33C/min up to 200C.
111
Chapter 6
In Figure 6.31 and Figure 6.32, heating with a ramp of 1 and 3C/min respectively, the same general behaviour in capacitance and tan is seen as in Figure 6.30. However, there is a rate dependency which is shown if looking at the reference point T(t=t2)=100C in time. The maximum value of tan for the lower frequencies is shifted towards longer times (higher temperatures).
Figure 6.31
Measured capacitance and tan for the resin-rich mica tape during a constant temperature ramp increasing from 20C with 1C/min up to 200C.
Figure 6.32
Measured capacitance and tan for the resin-rich mica tape during a constant temperature ramp increasing from 20C with 3C/min up to 200C.
In Figure 6.33, capacitance and tan are plotted at f=105 Hz for the three different ramps 0.33, 1, 3C/min as a function of time. The curves have all the same features but the maximum values for both capacitance and tan decreases with decreasing heating rate. The ridge in tan is also clearly seen at this frequency. In Figure 6.34, capacitance and tan are plotted at f=105 Hz for the three different ramps 0.33, 1, 3C/min as a function of temperature. This is a good way to plot the measurement data to illustrate the rate dependence as also seen above. The measured capacitance and tan values show similar behaviour up to 80C where the non-reversible and rate dependent curing process starts.
112

10-8
f = 105 Hz
Ramp 0.33C/min Ramp 1C/min Ramp 3C/min
Chapter 6
101
f = 105 Hz
10
-9
tan() 0 100 200 300 Time [min] 400 500 600
C [F]
10
10-10
10-1
100
200
300 Time [min]
400
500
600
Figure 6.33
10-8
Measured capacitance and tan at f=105 Hz during three different temperature ramps, 0.33, 1, 3C/min, as function of time.
101
f = 105 Hz
f = 105 Hz
10
-9
tan() 0 50 100 Temperature [C] 150 200
C [F]
10
10-10
10-1
50
100 Temperature [C]
150
200
Figure 6.34
Measured capacitance and tan at f=105 Hz during three different temperature ramps, 0.33, 1, 3C/min, as a function of temperature.
6.4.5.3
Curing under Factory temperature path
This resin-rich mica tape sample was pressed in the test cell, see chapter 5, with a constant pressure of 1.2 MPa and using a temperature path similar to what is used in the factory process, see Figure 6.35. The measurements started with the sample in the test cell at 20C and at t=t1 heating (9C/min) was done up to 100C, reached at t=t2. At t=t3 after around 15 min at 100C the temperature is increased with 4C/min up to 170C, reached at t=t4. At t=t5 after around 30 min at 170C the temperature is rapidly cooled to 100C. In the factory process, the stator bar is now ready to be lifted out of the press but here the sample is kept at 100C for 10 min and then cooled down to 20C before taken out of the test cell. An important difference between pressing insulation in the test cell and pressing a stator bar in the factory press is that the stator bar will contain all the copper conductors inside, which may result in different temperature distributions in the two cases.
113
Chapter 6
200 t4

Lower heating plate
Top heating plate Outside
Temperature [C]
150 t 100
2
t t
3
50
50 Time [min]
100
150
Figure 6.35
Measured temperature during curing of a resin-rich mica tape sample under a temperature path similar to the one used in the factory production.
In Figure 6.36, the capacitance and tan are measured during the temperature path described above, t=t1 to t=t6. For both the 100C and 170C temperature levels, it can be seen that capacitance and tan are decreasing during the time the temperature is constant. This is an effect from the cross-linking in the resin and the effect is stronger with temperature, which is also seen here. There are two mechanisms competing against each other, the dielectric response will increase with temperature but this will increase also degree of curing which will then decrease the dielectric response. The characteristic ridge in tan is also seen here and it passes through the whole measurement window before the measurement is ended. There is also a more distinct valley observed for the higher frequencies between t=t2 and t=t4, before the ridge is reached.
Figure 6.36
Measured capacitance and tan during curing under a similar temperature path as in the factory production.
In Figure 6.37, capacitance and tan for three frequencies, 50, 103 and 104 Hz, are shown as a function of press time. The same features as observed in Figure 6.35 above are also seen here. It is seen here that after t=t2, constant T=100C, both capacitance and tan decreases until t=t3. From t=t3 heating starts up to 170C, reached at t=t4, and both capacitance and tan increase slightly after which they decrease even though 170C is not reached. There is a peak following in tan, which is the ridge in Figure 6.36 passing by in time.
114

10
-8
Chapter 6
50 Hz 1 kHz 10 kHz
t1 t2
10
t
0
t3
t4 t5
50 Hz 1 kHz 10 kHz
t C [F]
-9
10
5
t6 tan() 10
-1
10
10-2
10-10
20
40
60 Time [min]
80
100
120
10-3
20
40
60 Time [min]
80
100
120
Figure 6.37
Measured capacitance and tan at 50, 103 and 104 Hz during a temperature cycle, t=t1 to t=t6, similar to the one used in the factory process.
After the sample was cured under the temperature path shown in Figure 6.35 it was cooled down and capacitance and tan were measured as a function of constant temperature, see Figure 6.38. The dielectric response is thermally activated in this temperature range with a single activation energy. There is also a high frequency loss peak, which is visible for the lower temperatures. One explanation might be that the high frequency loss peak represents the parts of the polymer chains that did not cross-link during curing and that this loss peak might change depending on the temperature path used for curing the sample. To get a better view of the capacitance measurements they are plotted on a semi-log scale compared to tan, which are plotted on a log-log scale.
9 10-10 10-1
48.0 C 29.0 C 7.4 C -9.1 C -28.4 C -47.7 C 48.0 C 29.0 C 7.4 C -9.1 C -28.4 C -47.7 C
8 10-10 tan() C [F] 10

-2
7 10-10
6 10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 10 10-1 100
101
102
103
104
105
106
Frequency [Hz]
Frequency [Hz]
Figure 6.38
After the sample was cured, under a similar temperature path as used in the factory process the capacitance and tan were measured for -50C<T<50C.
6.4.5.4
Curing under Soft temperature path
This resin-rich mica tape sample was pressed in the test cell, see chapter 5, with a constant pressure of 1.2 MPa and using a temperature path, see Figure 6.39, which is more continuous, more soft and slower in reaction rate than the temperature path used in the 115
Chapter 6
factory process. The measurements started with the sample in the test cell at 20C and at t=t1 heating (9C/min) was done up to 80C, reached at t=t2. At t=t2 the heating rate is decreased to 2C/min, heating the sample up to 160C, reached at t=t3. This temperature level was chosen since the resin-rich mica tape system is a class F (155C) system. At t=t4 after around 50 min at 160C the temperature is rapidly cooled to 100C. In the factory process, the stator bar is now ready to be lifted out of the press but here the sample is kept at 100C for 10 min and then cooled down to 20C before taken out of the test cell.
200 t Temperature [C] 150
3 Lower heating plate Top heating plate Outside
t4 t5
100
t2
50
50 Time [min]
100
150
Figure 6.39
Measured temperature during curing of a resin-rich mica tape sample under a soft (slower in time) temperature path compared to factory production.
In Figure 6.40, the capacitance and tan are measured during the soft temperature path described above, t=t1 to t=t5. What is seen in capacitance and tan for the lower frequencies is that their values are almost constant for about 16 min after t=t2, showing that the increase in temperature balances the rate of cross-linking in the resin.
Measured capacitance and tan during curing under a soft temperature path. Then there is a drop in capacitance and tan before 160C is reached at t=t3, showing at which time the degree of cross-linking starts to dominate over the temperature effect. The characteristic ridge in tan is also seen here and it passes through the whole measurement window before the measurement is ended and there is also here a distinct valley observed for the higher frequencies between t=t2 and t=t3, before the ridge is reached. Figure 6.40 116
Chapter 6
In Figure 6.41, capacitance and tan for three frequencies, 50, 103 and 104 Hz, are shown as a function of press time under the soft temperature path. The features described above are also seen here. The moving ridge is seen as a peak differently positioned in time depending on the measured frequency and the valley is seen for the higher frequencies.
10-8 t t2
1
10 t
3 50 Hz 1 kHz 10 kHz
t t2
1
50 Hz 1 kHz 10 kHz
t4 t5 10
-9
100 tan()
t4 t 5
C [F]
10
-1
10
-2
10-10 0 50 Time [min] 100 150
10
-3
50 Time [min]
100
150
Figure 6.41
Measured capacitance and tan at 50, 103 and 104 Hz during a soft temperature cycle, t=t1 to t=t5.
After the sample was cured under the soft temperature path shown in Figure 6.39 it was cooled down and capacitance and tan were measured as function of constant temperature, see Figure 6.42. The dielectric response seems to be thermally activated in this temperature range with a single activation energy.
9 10
-10 47.9 C 28.2 C 10.2 C -9.3 C -30.0 C -48.2 C
10
-1 47.9 C 28.2 C 10.2 C -9.3 C -30.0 C -48.2 C
8 10 C [F]
-10
tan()
-10 1 2 3 4 5 6
10-2
7 10
6 10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.42
Measured capacitance and tan for the fully cured sample, soft cure, as a function of frequency and temperature.
It is difficult to see any difference in this dielectric response compare to one measured after the factory like temperature path, see Figure 6.38. This indicates that with this measurement accuracy it might be difficult to say something about the way, the temperature paths used, to reach the final degree of curing just by measuring the final result.
117
Chapter 6
6.4.5.5
Curing under Hard temperature path
This resin-rich mica tape sample was pressed in the test cell, see chapter 5, with a constant pressure of 1.2 MPa and using a hard temperature path, see Figure 6.43, which is more discontinuous, more hard and faster in reaction rate than the temperature path used in the factory process. The measurements started with the sample in the test cell at 20C and at t=t1 heating (9C/min) was done up to 100C, reached at t=t2. At t=t3 after around 15 min at 100C the temperature is increased with 8C/min up to 190C, reached at t=t4. At t=t5 after around 5 min at 190C the temperature is rapidly cooled to 100C. The sample is kept at 100C for 10 min and then cooled down to 20C before taken out of the test cell. This temperature path is called hard because the manufacture of the resin-rich mica tape used does not recommend exceeding 180C when curing the tape.
200 t
4
t5
Lower heating plate Top heating plate Outside
Temperature [C]
150 t 100
2
t t
3
50
50 Time [min]
100
150
Figure 6.43
Measured temperature during curing of a resin-rich mica tape sample under a hard temperature path compared to the one used in the factory process.
In Figure 6.44, the capacitance and tan are measured during the hard temperature path described above, t=t1 to t=t6.
Measured capacitance and tan during curing under a hard temperature path. The measurement works well although strong temperature steps are made which could cause thermally induced noise. Similar behaviour as for the factory temperature path, see Figure 6.35, are also seen here, especially the characteristic ridge in tan. Figure 6.44 118
Chapter 6
In Figure 6.45, capacitance and tan for three frequencies, 50, 103 and 104 Hz, are shown as a function of press time under the hard temperature path. The features described above are also seen here and both capacitance and tan show the changes in temperature well. The moving ridge is seen as a peak differently positioned in time depending on the measured frequency.
10
-8
t1
t2
t3
10
4 50 Hz 1 kHz 10 kHz
t2
t3
50 Hz 1 kHz 10 kHz
10
t5
10
-9
tan() 0 20 40 60 Time [min] 80 100
C [F]
10
-1
10-2
10-10
10-3 0 20 40 60 Time [min] 80 100
Figure 6.45
Measured capacitance and tan at 50, 103 and 104 Hz during a hard temperature cycle, t=t1 to t=t6.
After the sample was cured under the hard temperature path shown in Figure 6.43, it was cooled down and capacitance and tan were measured as a function of constant temperature. In Figure 6.46, the dielectric response seems to be thermally activated in this temperature range with a single activation energy. Here it is also difficult to see any differences in the dielectric response compared to the responses measured after the factory like temperature path, see Figure 6.38, and the soft temperature path, see Figure 6.42.
9 10
-10 48.1 C 29.0 C 7.7 C -9.6 C -28.6 C -48.5 C
10
-1 48.1 C 29.0 C 7.7 C -9.6 C -28.6 C -48.5 C
8 10-10 C [F]
tan()
10
-2
7 10-10
6 10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.46
Measured capacitance and tan for the fully cured sample, hard cure, as a function of frequency and temperature.
119
Chapter 6
6.4.6
Mechanical measurements
It turns out to be difficult to see any differences in the dielectric response measured on fully cured insulation cured under the three different temperature paths discussed above see Figure 6.38, Figure 6.42 and Figure 6.46. An attempt to separate the three different temperature paths was to make a mechanical test of the adhesion between stand insulation and groundwall insulation. The basic idea was that the Soft cure temperature path, see Figure 6.39, would allow more cross-links to find each other whereas the Hard cure temperature path, see Figure 6.43, would freeze the system much faster and therefore isolate a few possible crosslinks. The number of cross-links will work as a binder both in between layers of insulation and towards the strands. Small simple test samples were designed, see Figure 6.47, in such a way that four layers of resin-rich insulation with woven glass as tape carrier were pressed together with 1 MPa in between two strands. The tape carrier material was always oriented in the same direction. A strand that was 10.7 mm wide and 1.9 mm thick was chosen and the length of the contact area was 50 mm giving a total length of the sample of 190 mm. The total length and thickness of the samples were important in order not to create stresses perpendicular to the tested surface. These unwanted stresses could appear if mounting a sample in the mechanical test bench in an incorrect way. First nine samples with the Soft cure temperature path and then ten samples with the Hard cure temperature path were pressed in the test cell. Two samples were pressed at the same time in order to keep the processing parameters as constant as possible, especially the applied pressure. The strands were first pre-cured in 180C for 1 hour to follow the real factory process as closely as possible. All ready pressed samples were stored in a refrigerator until performing the mechanical tests.
Top/Side view 10.7 mm
1.9 mm 50 mm 4 layers 190 mm
Figure 6.47
Design of the samples used for mechanical testing of the adhesion in between layers of insulation and towards the insulated strands.
The mechanical tests were made at room temperature in a test bench in which the sample was mounted vertically and calibrated so that the initial load force was around zero. The force was measured when pulling the sample apart with a speed of 1 mm/min. In Figure 6.48, the measured stress is plotted against extension for the samples cured under Soft (A) and Hard (B) temperature paths. Mechanical breakdown was defined as when the stress went down to 90% of the peak stress.
120

3.5 "Soft" curing 3 2.5 Stress [MPa] Stress [MPa] 2 1.5 1 0.5 0 0 0.2 0.4 0.6 Extension [mm] 0.8 1 3 2.5 2 1.5 1 0.5 0 0 0.2 0.4 0.6 3.5 "Hard" curing
Chapter 6
0.8
Extension [mm]
Figure 6.48
Measured stress plotted against extension for samples cured under (A) Soft and (B) Hard temperature paths.
What is seen in Figure 6.48 is that for both the Soft and Hard cured samples there is an elastic deformation in the beginning followed by a transition into a plastic deformation before total breakdown [92]. There are sometimes breakdowns, which are made in steps indicating that there is still some friction between the two surfaces of breakdown. This could be due to horizontal forces in the sample arising when putting the sample in the test bench. An observation is that all samples broke down between the strand and the tape side, which had the epoxy-resin/mica mixture on it (not the tape carrier side). There were mica flakes on each of the two sides of breakdown indicating that perhaps the breakdown started along the cleavage plane in the mica. In Figure 6.49, the mechanical breakdown probability, assuming a Weibull distribution, is plotted against breakdown stress (peak stress) for both the Soft cure and Hard cure temperature paths [13]. Both sets of breakdown data fit well to a Weibull distribution (goodness to fit >92%) see solid (Hard cure) and dotted (Soft cure) line.
0.999 0.99
Breakdown Probability
0.63 0.5 0.3 0.1 "Hard" cure Calc. curve "Hard" "Soft" cure Calc. curve "Soft"
0.01
0.001 1
Breakdown stress [MPa]
10
Figure 6.49
Mechanical breakdown probability plotted against peak breakdown stress for samples cured under Hard and Soft temperature paths using a Weibull distribution. 121
Chapter 6
Looking at the result there is a small tendency to a separation between the data from the Soft and the Hard cure temperature paths with a possible higher average breakdown stress for the Soft cure. However, there is no clear difference and more tests have to be done monitoring the curing parameters more carefully, especially the applied pressure, in order to confirm the separation between Soft and Hard cure temperature paths. The value at 0.1%, sometimes used as a dimensioning parameter, is found to be 1.2 MPa for both Soft and Hard cure. In Table 6.1, a summary is shown of the calculated statistical parameters from the two sets of Soft and Hard cure mechanical breakdown data. Soft cure
Samples Minimum [MPa] Maximum [MPa] Mean [MPa] Median [MPa] Std Deviation [MPa] Skewness (normal distribution) Kurtosis (normal distribution) 9 2.13 3.23 2.74 2.85 0.37 -0.23 -1.10
Hard cure
10 2.15 3.01 2.56 2.56 0.32 0.07 -1.57
Table 6.1
Calculated statistical parameters from mechanical breakdown tests made on samples cured under Soft and Hard temperature paths. Values refer to peak stresses.
The mean breakdown stress is 2.740.37 MPa for Soft cure compared to 2.560.32 MPa for Hard cure and the minimum measured breakdown stress is lower for Soft cure than for Hard cure. The parameters skewness and kurtosis describes the shape of the two measured distributions, Soft and Hard cure.
6.4.7
Factory measurement during pressing of stator bar
In co-operation with ABB Alstom Power, Vsters, Sweden the possibility was given to perform a full-scale on-line measurement when factory pressing a stator bar insulated with resin-rich mica tape insulation. The purpose was to measure the temperature gradient in the insulation and the change in capacitance and tan during the pressing stage. A test stator bar was manufactured with a main wall insulation thickness of d3.3 mm consisting of 18 layers half-lap, which were machine wrapped. When the tape was wrapped on to the stator bar, 17 thermocouples (Teflon insulated, dtc0.5 mm, Pentronic 04-20148, TT-40-K) were inserted according to Figure 6.50 at several different depths in the insulation, numbered from the copper conductor outwards: A1-A4, B1-B4, C1-C4, C top (top side) and C bottom (bottom side), D1-D2 and E. To measure the capacitance and tan a semiconductive tape was applied as a current electrode with a guard on the outside and all strands were connected as a voltage electrode. On the outside of the test stator bar an extra layer of half-lap Teflon tape was applied together with a polyester film (PET) to protect the press from liquid resin. The whole press was guarded and hats of aluminium foil were put on the ends of the test stator bar to reduce influence of external noise. 122

1700 mm
Chapter 6
200 Voltage electrode
500
200 D B
300
200 C
A Current electrode
500 E
HP 4284A LCR meter
HP 3852A Temperature logger
Figure 6.50
Schematic picture of the test stator bar positioned in the press. A, B, C, D and E indicate the positions of the thermocouples.
6.4.7.1
Temperature measurement
The test stator bar was put into the press at room temperature 19.3C, after which two horizontal pressings were made. The press cycle started, t=0 min, with manually controlling the heating of the press up to 100C. The press consisted of four individual heating/cooling elements named pressbox 1, 2, 3, 4. At t=25 minutes the press cycle was suppose to start but due to unexpected problems the start was delayed until t=47 minutes. The press cycle program started with a setpoint for 170C and at t=98 minutes that setpoint was changed to 100C and at t=108 minutes the program ended. Manual cooling was then performed down to 40C after which the test stator bar was left in the press for two hours before it was lifted out. In Figure 6.51, the results from the manually read temperature values of the four heating elements in the press are shown. The temperature values are the ones measured by the thermocouples in the press and are used in the control loop regulating the setpoint values.
200
Problem to start press cycle Pre-heating to T=100C
Press cycle
Press 1 Press 2 Press 3 Press 4
Temperature [C]
150
Manual cooling
100
50
50 Time [min]
100
150
Figure 6.51
Temperature values measured by the press control equipment from the four heating elements in the press, 1, 2, 3, 4.
123
Chapter 6
In Figure 6.52, the results from measuring the temperature at points A and B are shown. Thermocouples A2, A3 and B1 broke during assembly and therefore gave no information. At point A, there is a maximum value, 179C, reached for A4 (closest to the heating element) and around 8 min later there is a maximum value, 164C, reached for A1 (closest to the conductor), giving a Tmax=15C. In point B, the thermocouples B2, B3 and B4 follow each other closely reaching a maximum temperature value of 173C, Tmax1C.
200
Press cycle
A1 A2 (broken) A3 (broken) A4
200
Press cycle
B1 (broken) B2 B3 B4
150 Temperature [C]

Problem to start press cycle
150
Manual cooling
100
Pre-heating to T=100C
Temperature [C]
Manual cooling
100
50
50
0 0 50 Time [min] 100 150
0 0 50 Time [min] 100 150
Figure 6.52
(A) Temperature gradient at point A, thermocouples A2 and A3 are broken. (B) Temperature gradient at point B, thermocouple B1 broken.
In Figure 6.53, the results from measuring the temperature at points C, D and E are shown. At point C, there is a maximum value, 178C, reached for C4 (closest to the heating element) and around 7 min later there is a maximum value, 161C, reached for C1 (closest to the conductor), giving a Tmax=17C. The temperature at C bottom (bottom side) is around 4C higher than C top (top side) however they follow C1 well.
200
Press cycle
C1 C2 C3 C4 C bottom C top
200
Press cycle
D1 D2 E
Temperature [C]
Manual cooling
100
Pre-heating to T=100C
Temperature [C]
150
Problem to start press cycle
150
Manual cooling
100
50
50
0 0 50 Time [min] 100 150
0 0 50 Time [min] 100 150
Figure 6.53
(A) Temperature gradient at point C. (B) Temperature gradient at D and temperature at the end part of the test stator bar, E. At point D it is seen that D2 (closest to the heating element) reacts faster than any other thermocouple at a similar position to the pre-heating step to 100C. The difference between the maximum value at D1, 176C, and the maximum value at D2, 176C, is 4C with a time 124
Chapter 6
shift of around 8 minutes. Point E, which measures the temperature close to the conductor but outside the part of the test stator bar which is in the press, reaches a maximum temperature value of 117C.
6.4.7.2
Capacitance and tan measurement
At the same time as the temperature data at points A to E was measured the capacitance and tan were measured in the frequency range 25 Hz to 1 MHz with a measurement voltage of 9V. The final geometrical capacitance of the test stator bar is estimated to be C0280 pF. It is important to remember that a newly taped and unpressed stator bar contains a lot of air because of the rectangular geometry, which will affect the capacitance and tan. In Figure 6.54, the measured tan is shown throughout the whole time in the press: during manual preheating, press cycle program and manual cooling. There is a tendency, for the higher frequencies, towards an increase in tan, which perhaps can be related to the dielectric behaviour of the semiconductive tape that is used as a current electrode. The characteristic valley followed by the diagonally moving ridge in time and frequency are also seen between t=50 minutes and t=100 minutes indicating that the cross-linking in the resin is directing the insulation towards its final structure.
Figure 6.54
Measured tan for the test stator bar as a function of frequency and time, during manual preheating, press cycle program and manual cooling.
In Figure 6.55, the tan for three frequencies, 50, 103 and 104 Hz, are shown as function of press time. It can be seen that tan is increasing with temperature until around t65 minutes (for the 50 and 103 Hz curves) when the cross-linking in the resin decreases the tan value although the temperature is kept constant at 170C. At t=98 minutes cooling starts in the press cycle program down to 100C resulting in a drop in tan. An observation is that after the above mentioned decrease in tan there is a small peak appearing for each frequency curve f=50, 103, 104 Hz at times t=89, 77, 72 minutes.
125
Chapter 6
101
Pre-heating to T=100 C Problem to start press cycle
170C
50Hz 1 kHz 10 kHz
10 Tan
Manual cooling
10
-1
Press cycle
10
-2
50 Time [min]
100
150
Figure 6.55
Tan measured at 50, 103 and 104 Hz as a function of press time.
In Figure 6.56, the measured capacitance is shown throughout the whole time in the press: during manual preheating, press cycle program and manual cooling. What is seen is that during preheating the capacitance is increasing because the resin becomes liquid, behaving like a polar liquid, and that the geometry is changing because pressure is applied to the test stator bar. Approximately at the same time as tan starts to decrease the capacitance also starts to decrease. The cross-linking in the resin forms a bigger and bigger network that transforms the liquid resin into a solid dielectric binding together the mica flakes and woven glass. Around t=42 minutes there is a slight decrease in capacitance that is related to the problems starting the press cycle because restarting the press program resulted in loss of horizontal pressure.
Figure 6.56
Measured capacitance for the test stator bar as a function of frequency and time, during manual preheating, press cycle program and manual cooling.
In Figure 6.57, the capacitance for three frequencies, 50, 103 and 104 Hz, are shown as a function of press time. The capacitance drop around t=42 minutes due to loss of horizontal pressure is clearly seen especially for the higher frequencies.
126

10
-7 Problem to start press cycle Pre-heating to T=100C 50 Hz 1 kHz 10 kHz
Chapter 6
170C
Capacitance [F]
10
-8
Manual cooling
10
-9
Press cycle
10
-10
50 Time [min]
100
150
Figure 6.57
Capacitance measured at 50, 103 and 104 Hz as a function of press time.
6.4.7.3
Tip-up measurement after pressing the test stator bar
After the test stator bar was pressed, it was lifted out of the press to cool down to room temperature. The thermocouples that were pressed into the insulation were cut away as close to the insulation surface as possible and then the test stator bar was connected to a tan 50 Hz measurement bridge in ABB-Alstom Powers insulation laboratory for a tip-up measurement. Three different measurement sweeps were made, see Figure 6.58, with Umax=2, 13.8, 15 kV (rms) respectively. The accuracy of the bridge coupling for the lower voltages, around 200V, was not so good. For the last measurement at 15 kV (rms) there was PD activity which could come from the pressed in thermocouples. The free floating thermocouples were situated at least 200 mm outside the measurement electrode. The measured tan was around 0.011 (1.1%) for the higher voltages which is a bit high but still within the values given by the manufacturer (10 kV rms, 50 Hz, 23C tan: 0.8-1.2 %). The tape had been stored for some time and during wrapping of the test stator bar, the tape showed signs of being not so flexible.
10-1
f = 50 Hz
U U U
max max max
2kV 13.8kV 15kV
Tan
10
-2
10
-3
5 Voltage [kV]
10
15
Figure 6.58
Tip-up measurement at 50 Hz, Umax 2, 13.8 and 15 kV rms, measured at ABBAlstom Powers insulation laboratory in Vsters Sweden. Measurements were made at room temperature, T20C. 127
Chapter 6
6.4.7.4
Summary
The overall impression from the experiment of measuring both temperature and capacitance/tan at the same time, during pressing of a test stator bar was very positive. The temperature measurements gave good information about the temperature gradients in the insulation as a function of position in the press. The capacitance and tan measurements could be made with a low influence of external noise and the result reflects the status of the stator bar throughout the whole press procedure. Values of capacitance and tan will be a fingerprint of the production of this specific stator bar and can be stored and compared with other measurements from other stator bars. Capacitance and tan measurements made here show strong correlation to the simplified parallel-plate samples that have been pressed in the laboratory see chapter 6.4.5.3.
6.5
PET-film as carrier material
Another type of resin-rich mica tape was studied here that had polyester film (PET) as carrier material see Figure 6.59 [55]. The tape was impregnated with a modified high temperature epoxy novolac resin system which during curing when subjected to heat and pressure will first start to flow before gelling and then curing giving a homogenous void free insulation material.
Figure 6.59
Microscopic pictures of the resin-rich mica tape with polyester film (PET) as carrier material. There is no clear structure and the PET film will act as a homogenous barrier.
The epoxy-resin system in the tape is before curing in the B stage, giving a flexible tape with low volatile content. The fully cured insulation is a low loss and high temperature insulation system, which is designed for use in class F, 155C, type of rotating machines. In Figure 6.59, two microscopic pictures are shown from the tape showing that there is no clear structure and the PET film will act as a homogenous barrier restricting the flow of resin during curing. The uncured tape has a thickness of 0.120.02 mm and a total weight per unit area of 20020 g/m2. It is composed of 12012 g/m2 mica, 342 g/m2 polyester film (PET) and 505 g/m2 resin. After curing, the composite insulation has a density of 1.9 g/cm3, dielectric breakdown strength at 50 Hz is 43.5kV/mm (20C), dissipation factor is max 1.5% (20C, 0.4kV/mm, 50 Hz), relative permittivity is 4.5 (20C) and thermal conductivity is 0.22 W/mK. 128
Chapter 6
6.5.1
The uncured system
The dielectric response of the uncured resin-rich mica tape with polyester film (PET) as carrier material was measured in the frequency range 10 mHz to 1 MHz and in the temperature range -50C to 50C under a constant pressure of 1.2 MPa. The measurement series started at room temperature, then was cooled down in steps to -50C, then heated back up to room temperature, repeating that measurement, and then increased up to 50C. The maximum temperature was chosen under the assumption that the test sample would still be uncured after the measurement. In Figure 6.60, capacitance and tan are plotted showing that the uncured tape has a strong dispersion in both frequency and temperature.
Figure 6.60
Measured capacitance and tan for the uncured resin-rich mica tape with polyester film (PET) as carrier material as a function of frequency and temperature.
In Figure 6.61, capacitance and tan are plotted as a function of frequency for different temperatures. Strong temperature dependence in capacitance and tan is probably linked to the change in viscosity of the uncured tape.
10
-8 48.6 C 39.6 C 30.2 C 19.8 C 10.2 C 1.3 C -5.0 C -9.5 C -20.0 C -29.2 C -40.0 C -49.5 C
100
48.6 C 39.6 C 30.2 C 19.8 C
10.2 C 1.3 C -5.0 C -9.5 C
-20.0 C -29.2 C -40.0 C -49.5 C
10 10
-9
-1
tan() 10
-2
C [F]
10
-10
10-2 10-1 100
101
102 103
104 105
106
10
-3
10
-2
10
-1
10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.61
The measured capacitance and tan of the uncured resin-rich mica tape with polyester film (PET) as carrier material. The uncured tape is strongly temperature activated and for higher temperatures, there are two loss peaks entering the measurement window. 129
Chapter 6
There are two possible dipolar loss peaks in tan, which become visible in the measurement window for temperatures above 20C. It is difficult to say if the measured behaviour in capacitance and tan is due to the different materials in the composite or due to their geometrical configuration, see Figure 6.59. In this case, with a solid polyester film (PET) as carrier material it is possible that the second peak in tan is due to barrier effects from the film compared to the open structure of the woven glass carrier, see Figure 6.5.
6.5.2
The resin-rich mica tape with polyester film (PET) as carrier material was cured in the test cell, starting at 20C with rapid heating up to 80C and then increasing the temperature at 0.5C/min up to 160C keeping that temperature for 7 hours. This temperature path was chosen to be the same as used for curing the resin-rich mica tape with woven glass as carrier material. The temperature path follows the specifications given by the manufacturer also including a post-curing stage. The dielectric response of the fully cured tape was measured in the frequency range 10 mHz to 1 MHz and in the temperature range -50C to 200C under a constant pressure of 1.2 MPa. In Figure 6.62, the capacitance and tan are plotted as a function of frequency and temperature. The fully cured tape starts to show a strong dispersion in frequency for temperatures above 70C. For the lower temperatures, the dispersion in the measurement window is weaker and there is a visible minimum in tan.
Figure 6.62
Measured capacitance and tan for the fully cured resin-rich mica tape with woven glass as carrier material as a function of frequency and temperature.
In Figure 6.63, capacitance and tan are plotted as function of frequency for different temperatures. For temperatures higher than 130C there is a peak in tan visible in the measurement window which is probably one of the two peaks seen for the uncured tape, see Figure 6.61, but shifted to lower frequencies (higher temperature) as the taped is cured. The dielectric responses for resin-rich mica tapes of the two types studied here, woven glass and polyester film as carrier material, look more the same in the fully cured case, see Figure 6.7, than they do in the uncured case, compare Figure 6.5 and Figure 6.61. This indicates perhaps that structure of the tape is more important in the uncured case when the viscosity of the resin is lower and the conductivity higher compared to the fully cured case.
130

10-8
200 C 170 C 150 C 130 C 110 C 90 C 70 C 50 C 30.4 C 20.0 C 10.1 C 0.2 C -10.7 C -20.7 C -29.9 C -39.5 C -49.5 C
Chapter 6
10
0 200 C 170 C 150 C 130 C 110 C 90 C 70 C 50 C 30.4 C 20.0 C 10.1 C 0.2 C -10.7 C -20.7 C -29.9 C -39.5 C -49.5 C
10-1 tan() 10-2 106
C [F]
10
-9
10
-10
10-2 10-1 100
101
102 103
104 105
10-3 10-2 10-1 100 101 102 103 104 105 106 Frequency [Hz]
Frequency [Hz]
Figure 6.63
The measured capacitance and tan of the fully cured resin-rich mica tape. There is a loss peak in tan entering the frequency window for temperature higher than 100C.
6.5.3
In order to investigate the behaviour of the dielectric response for different degrees of curing four resin-rich mica tape samples with polyester film (PET) as carrier material were cured under different temperature paths but with constant pressure, 1.2 MPa. After each profile, a DSC measurement was made for each sample to determine the degree of curing. The results from the DSC measurements, total remaining exothermal energy QT, were then compared to the total exothermal energy, Qtot, in the uncured tape. The first step before curing each of the four samples was to make DSC measurements on the uncured resin-rich mica tape. In Figure 6.64, the results are shown from two DSC measurements, one with the PET film carrier material and one with the carrier material removed which was done by first cooling down the tape and then bending away the PET film.
0.4
0.4
0.3
0.3
0.2
0.2
0.1
Q 48.2 [J/g]
tot
0.1
Qtot 44.1 [J/g]
0 50
100
150
200
250
300
0 50
100
150
200
250
300
Temperature [C]
Temperature [C]
Figure 6.64
DSC measurements on uncured resin-rich mica tape with (A) and without (B) PET tape carrier material. The total exothermal energy was Qtot46 J/g. 131
Chapter 6
The PET film contributes with an endothermal peak at around 250C, fortunately this endothermal peak is rather well separated from the exothermal peak making it still possible to estimate the reaction energy in the resin. The total exothermal energy per gram of the resinrich mica tape was estimated from a scaled baseline, taking away the endothermal peak, to be Qtot46 J/g. The baseline was scaled, meaning that it was estimated from the heat capacities for the uncured and cured state and in between scaled with the degree of curing. Four samples were then cured according to the temperature paths shown in Figure 6.65 (A). The temperature paths start at 20C, heating rapidly up to 80C and then changing heating rate to 0.5C/min up to Tmax=116, 126, 136, 151C respectively. Cooling was then done at 0.5C/min down to 80C and then maximum cooling down to 10C. These profiles were chosen more or less on a trial and error basis and they are the same as used for the resin-rich mica tape with woven glass as carrier material. In Figure 6.65 (B), the DSC measurements are plotted for the uncured tape and for the four samples, 2 to 5, cured according to the temperature paths described above. There is a clear shift and reduction in exothermal energy with increasing degree of curing. The endothermal peak from the PET film is unchanged for all samples and from this the assumption is made that the PET film is unaffected of the curing process. The degree of curing was then calculated for Tmax=116, 126, 136, 151C to be 0.1, 0.3, 0.5, 0.8.
160 140 Temperature [C] 120 100 80 60 40 20 0 0 50 100 150 200 250 300 350 Time [min] 0 50 100 150 200 250 300 Temperature [C] (2) (3) (4) (5) 151 C 136 C 126 C 116 C 0.4
0.3
0.2
4 3
0.1
1
Figure 6.65
(A) The different temperature paths used to cure the samples of resin-rich mica tape with PET film as carrier material. (B) Measured DSC for the partially cured samples and from that the degrees of curing were calculated.
6.5.3.1
The capacitance and tan were measured for the sample that was cured to 0.1 in the frequency range 10mHz to 1 MHz changing the temperature from 10C to 50C. In Figure 6.66, measured capacitance and tan are plotted against frequency for different temperatures. The dielectric response is very similar in shape and magnitude to the one measured for the uncured tape, see Figure 6.61, with the difference that the two loss peaks for the higher temperatures are here shifted one decade towards lower frequencies. In this temperature region, there seems to be a single activation energy describing the temperature dependence.
132

10
-8 50 C 40 C 30 C 20 C 10 C
Chapter 6
10
0 50 C 40 C 30 C 20 C 10 C
10-1 10-9 tan() 10-2

1 2 3 4 5 6
C [F]
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.66
Measured capacitance and tan for a resin-rich mica tape sample with PET film as carrier material cured to 0.1. The two peaks in tan are shifted towards lower frequencies compared to 0.
6.5.3.2
The capacitance and tan were measured for the sample that was cured to 0.3 in the frequency range 10mHz to 1 MHz changing the temperature from 10C to 50C. In Figure 6.67, measured capacitance and tan are plotted against frequency for different temperatures. The dielectric response is also here shifted one decade towards lower frequencies compared to the tape cured to 0.1, see Figure 6.61. Only one loss peak is now seen in the measurement window for temperatures above 30C, the first one is at this degree of curing shifted out of the frequency window. For this degree of curing, 0.3, and in this temperature region, it seems to be a single activation energy describing the temperature dependence.
10
-8 50 C 40 C 30 C 20 C 10 C
10
0 50 C 40 C 30 C 20 C 10 C
10 10-9 tan() 10 C [F]
-1
-2
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10
-3
10
-2
10
-1
10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.67
Measured capacitance and tan for a resin-rich mica tape sample with PET film as carrier material cured to 0.3. There is only one peak left in tan in this measurement window compared to 0.1 since the other one has been shifted out towards lower frequencies.
133
Chapter 6
6.5.3.3
The capacitance and tan were measured for the sample that was cured to 0.5 in the frequency range 10mHz to 1 MHz changing the temperature from 10C to 50C. In Figure 6.68, measured capacitance and tan are plotted against frequency for different temperatures. The two loss peaks, which are seen for the lower degrees of curing, are here almost totally shifted out of the measurement window. There is also a possible high frequency loss peak seen in tan. The temperature dependence is stronger for the higher temperatures and there is a change in activation energy around 20C.
10
-8 50 C 40 C 30 C 20 C 10 C
10
0 50 C 40 C 30 C 20 C 10 C
10 10-9 tan() 10 C [F]
-1
-2
10-10 -2 -1 0 10 10 10
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.68
Measured capacitance and tan for a resin-rich mica tape sample with PET film as carrier material cured to a0.5.
6.5.3.4
The capacitance and tan were measured for the sample that was cured to 0.8 in the frequency range 10mHz to 1 MHz changing the temperature from 10C to 50C. In Figure 6.69, capacitance and tan are plotted as function of frequency for different temperatures.
10
-8 50 C 40 C 30 C 20 C 10 C
10
-1 50 C 40 C 30 C 20 C 10 C
10-9
tan()
-2 -1 0 1 2 3 4 5 6
C [F]
10
-2
10
-10
10
10
10
10
10
10
10
10
10
10-3 -2 -1 0 10 10 10
10
10
10
10
10
10
Frequency [Hz]
Frequency [Hz]
Figure 6.69
Measured capacitance and tan for a resin-rich mica tape sample with PET film as carrier material cured to a0.8. 134
Chapter 6
The two loss peaks, which are seen for the lower degrees of curing, are here totally shifted out of the measurement window and the high frequency loss peak seen in tan is more significant. For this degree of curing, 0.8, and in this temperature region, it seems to be a single activation energy describing the temperature dependence. The activation energy for the differently cured samples seems to be decreasing with increasing degree of curing, .
6.6
Summary
In this chapter, two different resin-rich mica tapes were studied. The fist tape had woven glass as carrier material and the second had polyester film (PET). For the woven glass tape, measurements were made at different degrees of curing and at constant temperature in the range -50C to 50C. In the tan part (loss part) of the dielectric response there is a clear shift of the response at a constant temperature from higher towards lower frequencies as the degree of curing increases. The magnitude of tan decreases in the studied frequency window, 10 mHz to 1 MHz, rapidly in the beginning up to a degree of curing of 0.6 after which it flattens out. This is interpreted as that the final structure of the epoxy-resin network is reached around 0.6 and that for higher degrees of curing more cross-links are added only to make the structure more rigid. There is also a tendency to a high frequency peak, >1 MHz, which is seen for the lower temperatures, <0C. This peak is also shifted towards lower frequencies as degree of curing increases. For the resin-rich mica tape with woven glass as carrier material, the dielectric response was also measured parallel to the surface of the tape. In this direction, the tape shows a stronger dispersion in both capacitance and tan and a higher magnitude of loss for both the uncured and fully cured cases. The dielectric response was also measured for the resin-rich mica tape with woven glass during different temperature paths. Fundamental features for all temperature paths is than an increase in temperature will lead to an increase in dielectric response but as temperature increases degree of curing will also increase which as a secondary effect will decrease the dielectric response. These two effects work against each other. Looking at dielectric response measurements that were made during curing under temperature ramps with different slopes they show a clear rate dependence. It was assumed that the rate of cross-linking would affect the final quality of the insulation in terms of built in tension and adhesion between layers and towards the conductor. To verify this three different temperature paths were chosen; Factory, Soft and Hard, that would give the same final degree of curing. The Factory path simulated the factory process. The Soft path had slower changes in the temperature and lower maximum value, which would give a lower and more constant (in time) rate of cross-linking. The Hard path had faster changes in temperature and higher maximum value, which would quickly freeze the system isolating possible cross-links. The measured dielectric response during the three temperature paths showed similar features, especially a ridge in tan that moved through the measurement window. However, mechanical breakdown test of the adhesion between layers and towards the conductor showed only a small tendency towards a higher breakdown stress for the Soft temperature path. The assumption made about the relation between rate of cross-linking and mechanical properties could not be verified.
135
Chapter 6
The dielectric response and the temperature gradient were measured during pressing in the factory process of a stator bar wrapped with resin-rich mica tape with woven glass as carrier material. The dielectric response showed the same features, especially the moving ridge in tan, as the parallel-plate samples pressed in the laboratory. Capacitance and tan measurements gave a finger print of the pressing/processing stage of the stator bar. Features like loss of horizontal pressure in the press could be identified in the capacitance measurement. The second resin-rich mica tape had a polyester film (PET) as carrier material. Also for this tape, the dielectric response was measured at different degrees of curing and at constant temperature in the range 10C to 50C. For the lower degrees of curing two peaks were seen in tan compared to one for the tape with woven glass. This second peak believed to originate from the barrier effect due to the solid PET film. When the degree of curing is increased, the dielectric response is shifted towards lower frequencies. Differential scanning calorimetry (DSC) measurements show clearly a peak (endothermal energy) related to melting of the PET film. However, the exothermal peak from the epoxyresin and the endothermal peak from the PET film is separated enough in temperature that it is still possible to estimate degree of curing from a DSC measurement. There is no corresponding endothermal peak for the tape with woven glass as carrier material.
136
Modelling a resin-rich insulation system
Chapter 7
7.1
Introduction
In this chapter a simple Arrhenius activated reaction model will be used to describe the progress of curing, change in , for the two resin-rich mica tapes studied in this thesis. The parameters in the reaction model will be estimated from differential scanning calorimetry (DSC) measurements. When knowing these parameters the degree of curing can be calculated for an arbitrary temperature path which makes it possible to relate degree of curing, , to the measured dielectric response, capacitance and tan. The master curve approach will be used here to study the temperature dependence, thermal activation, and how this is changing with degree of curing, . It is expected that the thermal activation will be reduced with increasing degree of curing, which relates both to the change in viscosity and glass transition temperature during the curing process. To explain the dielectric response from the resin-rich mica tapes a circuit network model is proposed consisting of the materials building up the tapes. Mica and the two tape carrier materials woven glass and polyester (PET) are supposed to be unchanged during the curing process and will only have a temperature dependence. The dielectric response of the resin will change with both degree of curing, , and temperature. Certain features like barrier effects can be explained with the geometrical configuration of the materials in the network.
7.2
Modelling the curing of a resin
The curing reaction of a resin is accompanied by changes in a variety of physical properties like density, electric conductivity, refractive index, viscosity and dynamic mechanical properties. To understand and to be able to model the structural and physical changes that follow cross-linking in a resin it is necessary to have basic understanding of the kinetics controlling the cure reaction. Since most curing reactions in resins are of exothermal nature, it is of need to have knowledge about rate of heat generation and heat flow, as a function of time and temperature history experienced by the resin [100]. The total heat flow to a resin sample subjected to a time dependent temperature T=T(t) can be written as
dQ = m C p (T, dt
) dT
dt
Exothermal heat flow
d Q tot dt 1 4 24 3
(7.1)
where m is mass, Cp is specific heat, Qtot total exothermal energy in the reaction and is degree of curing. It is generally assumed that the exothermal energy released in the reaction is proportional to the number of bonds that have reacted in the system. Because of this, the 137
Chapter 7
cumulative heat evolved at any time will be related to the extent of reaction at that time. Based on this the degree of curing at any time t can be written as
(t ) = 1 Q T (t )
Q tot
(7.2)
where QT(t) is the remaining exothermal energy at time t. What is not taken into consideration here is that the maximum degree of curing is actually dependent on temperature, in many cases linearly [70]. This result in that Qtot actually represents the total exothermal energy released using sufficiently high temperature so that the resin is fully cured. The simplest kinetic model used to describe curing under an arbitrary temperature T=T(t) is the nth order rate expression [59], [65], [100]
E d T (t ) (1 = Ae dt (t = 0) = 0
)n
(7.3)
where A, E and n are reaction rate parameters. This first order differential equation can be solved directly and the solution is formed as
t E 1 n (t ) = 1 1 A (1 n ) e T ( t ) dt 0 1
n 1
(7.4)
The kinetic model described in equation (7.3) is used throughout this thesis because of its simplicity but to illustrate that there are also other possibilities the degree of curing can be written as [65], [100]
E E d 1 2 T (t ) T (t ) + A2 e = A1 e dt (t = 0) = 0 m
(1
)n
(7.5)
This expression is considered to fit experimental data more exactly but it does not change the basic ideas behind the approach. The total heat flow to a resin sample under a given temperature ramp is measured in a differential scanning calorimetry (DSC) measurement. From the measured heat flow, it is possible to estimate the total exothermal energy Qtot and the reaction parameters A, E and n in equation (7.3). To do this it is first necessary to find the baseline, which is the heat flow to the resin due to its specific heat. The specific heat is changing both with temperature and degree of curing but the two effects can be separated by assuming the following C p (T,
) = C p (T,
= 0 ) (1 14243
C1 + C2 T
) + C p (T,
= 1) 14243
C3 + C4 T
(7.6)
where the specific heat at low and high degrees of curing is modelled with a linear temperature dependence. The specific heat mentioned here is for the whole resin-rich mica tape with woven glass or polyester film (PET) as carrier material. Since there is a spread in the composition of materials in the tape, the baseline will change accordingly and therefore
138
Chapter 7
also indirectly the estimations of Qtot, A, E and n. From equations (7.1) and (7.6) the total exothermal energy Qtot can now be estimated as Q tot = Q tot (T(t end )) =
t end
m C p (T,
) dT dt
dt Measured DSC dQ dt {
(7.7)
where T=T(tend) is the end temperature of the DSC measurement which was in all experiments 300C, a temperature at which the resin is fully cured. Knowing Qtot it is also possible, using the same baseline as in equation (7.7) to calculate the change in degree of curing under the DSC measurement as follows
t 1 (t ) = m C p (T, Q tot 0
) dT dt
dt Measured DSC dQ dt {
(7.8)
When Qtot and a(t) are known they can be put into equations (7.1) and (7.3) giving
1 m C p (T, Q tot m C p (T, X (t ) = ln Q tot
) dT dt
E d T (t ) = Ae (1 = dt Measured DSC dQ dt {
)n
dT 1 dt Q tot
E + n ln (1 = ln (A ) T (t ) Measured DSC dQ dt {
(t ))
(7.9)
1 1 T (t 0 ) ln (1 (t 0 )) ln (A ) X (t 0 ) M E = M M M ( ( ) ) ( ) ( ) 1 1 T t ln 1 t n X t m m m
The reaction rate parameters A, E, and n can then be estimated by taking the natural logarithm of both left and right sides and then solving the set of linear equations received when stepping through time. To get best least-square fits the time vector is centered around the exothermal heat flow peak.
7.2.1
Resin-rich mica tape with woven-glass as carrier material
Several DSC measurements were done on the uncured resin-rich mica tape with woven glass as carrier material to determine the reaction parameters, A, E and n in equation (7.3). The DSC measurements show good reproducibility, see Figure 6.8, and a typical measurement is shown below in Figure 7.1. A scaled baseline, see equation (7.6), is used and the reaction parameters are estimated from a total of six measurements giving the results of total reaction energy Qtot=14816 J/g and reaction parameters A=exp(26.131.90) 1/s, E=(1.340.07)104 K (1.15 eV) and n=2.270.26.
139
Chapter 7
0.6
M-T, m=8.73 m g Scaled baseline
0.4 Heat flow per gram [W /g]
0.2
0
Q = 148+/-16 [J/g]
tot
-0.2
A = exp(26.13+/-1.90) E = (1.34+/-0.07) 10 [K] n = 2.27+/-0.26

4
-0.4 50 100 150 200 250 300 Temperature [ C]
Figure 7.1
Typical DSC measurement, heating rate 9C/min, on uncured resin-rich mica tape with woven glass as carrier material.
DSC measurements were also done with other heating rates in order to investigate how the rate dependence would affect the reaction parameters in equation (7.3). In Figure 7.2, heating rates were chosen to (A) 3C/min and (B) 1C/min. To estimate the reaction parameters turns out to be a numerically sensitive procedure and therefore the part of the DSC curve used for estimation was defined as the minimum heat flow 85% of start and end heat flow values. What is seen when comparing Figure 7.1, 9C/min, with Figure 7.2, 3C/min and 1C/min, is that the total reaction energy Qtot decreases with decreasing heating rate. If the exothermal energy given away during the chemical reaction is proportional to the number of cross-links generated, the total number of cross-links will decrease with decreasing heating rate. The reaction rate parameters also change but their change are more uncertain and more experiments are needed before any further conclusions can be made.
0.2 M-T, m=3.11m g Scaled baseline 0.15 Heat flow per gram [W /g] Heat flow per gram [W /g] 0.15 0.2 M-T, m=2.66 mg Scaled baseline
0.1
0.1
0.05 Q 0
tot
= 135 [J/g]
4
Q 0.05
tot
= 108 [J/g]
4
A = exp(23.55) E = 1.23 10 [K] n = 1.95
A = exp(26.36) E = 1.35 10 [K] n = 1.85
-0.05 50 100 150 200 250 300 Temperature [ C]
0 50 100 150 200 250 300 Temperature [ C]
Figure 7.2
Two DSC measurements with different heating rates, (A) 3C/min and (B) 1C/min. The total exothermal reaction energy Qtot decreases with decreasing heating rate.
140
Chapter 7
7.2.1.1
When the reaction rate parameters in equation (7.3) are estimated, the set of 9C/min DSC measurements are used for the estimation, the degree of curing (t) can be calculated for any given temperature path T=T(t). In Figure 7.3 (A) the measured temperature paths from curing the resin-rich mica tape samples with woven glass as carrier material to different degrees are plotted and in (B) the progress of the degree of curing for the different temperature paths are calculated.
160
Tmax = 151 C
1
Tmax = 151 C Tmax = 136 C
140 120 Temperature [ C] 100 80 60 40 20 0 0
Calculated degree of curing,
Tmax = 126 C Tmax = 116 C
0.8
Tmax = 136 C
0.6
Tmax = 126 C
0.4
Tmax = 116 C
0.2
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350 Time [min] Time [min]
Figure 7.3
(A) Temperature paths used for curing the resin-rich mica tape to different degrees. (B) Calculated degree of curing as a function of time for samples cured under the temperature paths in (A).
In Figure 7.4 (A), DSC measurements are shown from samples cured under the temperature paths found in Figure 7.3 (A). From these measurements, the degree of curing for each sample can be calculated by relating the measured exothermal energy to the total exothermal energy for the uncured tape.
0.6
1
5
Measured, DSC, degree of curing,
0.4 Heat flow per gram [W/g]
0.8
3
0.2
4 3
0.6
2
0
1
0.4
-0.2
0.2
-0.4 50 100 150 200 250 300 Temperature [C]
0 0 0.2 0.4 0.6 0.8 1 Calculated degree of curing,
Figure 7.4
(A) DSC measurements for samples with different degrees of curing. From these measurements, the degree of curing is calculated. (B) Measured degree of curing compared to calculated degree of curing. Two samples for each degree of curing. 141
Chapter 7
Two DSC measurements, from two different positions, on each sample were made and in Figure 7.4 (B) the measured degree of curing is plotted against the calculated degree of curing. The correlation between measured and calculated degree of curing is good.
7.2.1.2
Curing under constant temperature
Resin-rich mica tape samples with woven glass as carrier material were cured in the test cell under constant temperature, see temperature paths in Figure 7.5 (A). Using the proposed reaction rate expression in equation (7.3) with estimated parameters, see Figure 7.1 the development in degree of curing can be calculated as in Figure 7.5 (B).
200 p 6 kPa 180 C Calculated degree of curing, (t) 160 C 150 Temperature [C] 140 C 120 C 100 0.8 1 180 C 160 C 140 C
0.6
120 C
0.4
50
0.2
0 0 50 100 150 200 250 Time [min]
0 0 50 100 150 200 250 Time [min]
Figure 7.5
(A) Temperature paths used for curing under constant temperature. (B) Calculated degree of curing for the temperature paths in (A).
Capacitance and tan were measured during the temperature paths in Figure 7.5 (A), see chapter 6.4.5.1. In Figure 7.6 tan at f=50 Hz is plotted against calculated degree of curing under constant temperatures T=120, 140, 160, 180C.
100 f = 50 Hz
10 tan() 140 C 160 C 1 120 C 0.1 0 0.2 0.4 0.6 0.8 1 Calculated degree of curing, 180 C
Figure 7.6
Measured tan at f=50 Hz as a function of degree of curing under constant temperatures T=120, 140, 160, 180C.
142
Chapter 7
The behaviour of tan for each temperature is similar but they are shifted along the degree of curing axis. It is important to remember that the maximum degree of curing here is assumed independent of temperature. This together with the fact that tan for the uncured part, 1-, is shifted in frequency with temperature will probably re-normalise the curves in Figure 7.6.
7.2.1.3
Curing under temperature ramp
Resin-rich mica tape samples with woven glass as carrier material were cured in the test cell under constant temperature ramps, see temperature paths in Figure 7.7 (A). Using the proposed reaction rate expression in equation (7.3) with estimated parameters, see Figure 7.1 the development in degree of curing can be calculated as in Figure 7.7 (B).
250 Calculated degree of curing, (t) 3 C/min 1 C/min 200 Temperature [C] 0.33 C/min 1
0.8 0.33 C/min 0.6 3 C/min
150
1 C/min 0.4
100
50 p 6 kPa 0 0 100 200 300 Time [min] 400 500 600
0.2
0 0 100 200 300 Time [min] 400 500 600
Figure 7.7
(A) Temperature paths used for curing under constant temperature ramps. (B) Calculated degree of curing for the temperature paths in (A).
Capacitance and tan were measured under the temperature paths in Figure 7.7 (A), see chapter 6.4.5.2. In Figure 7.8 tan at f=105 Hz is plotted against calculated degree of curing for curing under constant temperatures ramps 0.33, 1, 3C/min.
10 f = 105 Hz 1 C/min 3 C/min tan() 1 0.33 C/min
0.1 0 0.2 0.4 0.6 0.8 1 Calculated degree of curing,
Figure 7.8
Measured tan at f=105 Hz as a function of degree of curing for curing under constant temperatures ramps, 0.33, 1, 3C/min. 143
Chapter 7
The behaviour of tan for each temperature ramp is similar but they are also, like in Figure 7.6, shifted along the degree of curing axis. The different degrees of curing are also here reached at different temperatures, shifting tan in frequency since it has temperature dependence. In Figure 7.1 and Figure 7.2 it is seen that the total exothermal reaction energy decreases with decreasing heating rate which can change reaction rate parameters and therefore indirectly also the degree of curing.
7.2.1.4
Curing under Factory, Soft and Hard temperature paths
Resin-rich mica tape samples with woven glass as carrier material were cured in the test cell under three different temperature paths called Factory, Hard and Soft paths, see Figure 7.9 (A). These three temperature paths were chosen to represent different ways to cure the insulation. Hard curing is a rapid process with high temperatures compared to Soft curing which is a slower process with lower temperatures. Using the proposed reaction rate expression in equation (7.3) with estimated parameters, see Figure 7.1 the development in degree of curing can be calculated as in Figure 7.9 (B).
200 "Hard" Factory Calculated degree of curing, (t) "Soft" 150 Temperature [C] 0.8 p 1.2 MPa 1 "Hard" "Soft" Factory
0.6
100
0.4
50
0.2
0 0 50 Time [min] 100 150
0 0 50 Time [min] 100 150
Figure 7.9
(A) Temperature paths used for curing under three different temperature paths, Factory, Hard and Soft. (B) Calculated degree of curing for the temperature paths in (A).
Capacitance and tan were measured under the temperature paths in Figure 7.9 (A), see chapter 6.4.5.3 to 6.4.5.5. In Figure 7.10, Factory temperature path, and in Figure 7.11, Soft and Hard temperature paths, tan at f=50, 103, 104 Hz are plotted against calculated degree of curing. The general behaviour for tan in both figures, three temperature paths, are the same. The peaks in tan for >0.5 are seen for all three temperature paths and they are shifted towards lower degrees of curing with higher frequencies. Following the peak, there is also for all temperature paths a steep drop in tan. The temperature step up to 100C for the Factory and Hard temperature paths is clearly seen as a dip in tan around 0.1. There are two possibilities here to monitor and diagnose the curing process of the insulation. The first is to measure the temperature on each side of the insulation and then if there is a small temperature gradient calculate the degree of curing. The second is to measure the change in capacitance and tan, relating for example the peak in tan for f=50 Hz, assuming Factory temperature path, to a degree of curing >0.9. Changes in capacitance and tan will 144
Chapter 7
also reveal disturbances in the process like sudden changes in temperature or pressure, see also Figure 6.54 and Figure 6.56.
10 Factory f = 50 Hz 1 tan() f = 1 kHz
0.1
f = 10 kHz
Figure 7.10
10
Measured tan at f=50, 103, 104 Hz as a function of degree of curing under a temperature path similar to the one used in the factory process.
10 "Soft" f = 50 Hz f = 50 Hz "Hard"
1 tan() tan() f = 1 kHz
1 f = 1 kHz
0.1
f = 10 kHz
0.1
f = 10 kHz
Figure 7.11
Measured tan at f=50, 103, 104 Hz as a function of degree of curing under (A) a Soft temperature path and (B) a Hard temperature path.
7.2.2
Resin-rich mica tape with polyester film as carrier material
DSC measurements were done on the uncured resin-rich mica tape with poyester film (PET) as carrier material to determine the reaction parameters, A, E and n in equation (7.3). In Figure 7.12 (A), a DSC measurement with PET film is shown and in Figure 7.12 (B), the PET is removed before making the measurement. A scaled baseline, see equation (7.6), is used and fit to the measurement is such away that the endothermal peak related to the PET film is removed. The total exothermal energy is estimated to Qtot48 J/g and the reaction parameters to Aexp(23), E1.3104 K and n1.7. The total exothermal energy is smaller in this system, 145
Chapter 7
less epoxy-resin, compared to the resin-rich mica tape with woven glass as carrier material, which is also consistent with tape specifications in chapter 6.4 and 6.5.
0.4
Q 48.2 [J/g]
tot
0.4
A exp(23.5) M-T, m=7.48 mg Scaled baseline Q 44.1 [J/g]
tot
A exp(21.0)
n 1.7
0.3
E 1.3 10 [K]
0.3
E 1.2 10 [K] n 1.7
0.2
0.2
0.1
0.1
0 50
100
150 200 Temperature [C]
250
300
0 50
100
250
300
Figure 7.12
Typical DSC measurements, heating rate 9C/min, on uncured resin-rich mica tape (A) with polyester film (PET) as carrier material and (B) with PET film taken away.
7.2.2.1
When the reaction rate parameters in equation (7.3) are estimated the degree of curing (t) can be calculated for any given temperature path T=T(t). In Figure 7.13 (A), the measured temperature paths are shown used for curing the resin-rich mica tape samples with polyester film (PET) as carrier material to different degrees. In Figure 7.13 (B) the corresponding progress in the degree of curing for the different temperature paths are calculated. The same temperature paths are used as for the curing of the resin-rich mica tape with woven glass as carrier material but here different degrees of curing, lower, are reached.
160
Tmax = 151 C
1
Tmax = 136 C Tmax = 126 C Tmax = 116 C
140 120 Temperature [ C] 100 80 60 40 20 0 0
0.8
max
= 151 C
0.6
T
max
= 136 C
0.4
T
max
= 126 C
0.2
max
= 116 C
0
50 100 150 200 250 300 350
50
100
Time [min]
150 200 Time [min]
250
300
350
Figure 7.13
(A) Temperature paths used for curing the resin-rich mica tape to different degrees. (B) Calculated degree of curing as a function of time for samples cured under the temperature paths in (A).
146
Chapter 7
In Figure 7.14 (A), DSC measurements are shown from samples cured under the temperature paths found in Figure 7.13 (A). From these measurements, the degree of curing for each sample can be calculated by relating the measured exothermal energy to the total exothermal energy for the uncured tape when the endothermal peak for the PET film is taken away.
0.4
Measured, DSC, degree of curing,
0.8
0.3
0.6
0.2
4 3
0.4
0.1
1
0.2
0 50
100
250
300
0.2
0.4
0.6
0.8
Figure 7.14
(A) DSC measurements for samples with different degrees of curing. From these measurements, the degree of curing is calculated. (B) Measured degree of curing compared to calculated degree of curing.
One DSC measurement on each sample were made and in Figure 7.4 (B) the measured degree of curing is plotted against the calculated degree of curing. The correlation between measured and calculated degree of curing is good.
7.3
Temperature dependence for the resin-rich mica tapes
The temperature dependence for the resin-rich mica tapes with woven glass and polyester film (PET) as carrier material can be studied using the master curve shifting technique discussed in chapter 3.10. The basic idea is that the dielectric response is shifted, on a log-log scale, in frequency and amplitude when changing the temperature rather than that the shape of the dielectric response changes. This can be expressed in the following way
( ,T, = A y (T, Tref ) ( A x (T, Tref ) , Tref ,
(7.10)
where (,T,) is the complex electric susceptibility, degree of curing and T and Tref representing the shift in temperature. The factor Ax,y(T,Tref) can be expressed in different ways but a common expression is to use the Arrhenius factor A x,y (T, Tref ) = e
E x, y 1 1 kB T Tref
(7.11)
where kB is Boltzmanns constant. The activation energy Ex,y can be dependent on temperature and degree of curing.
147
Chapter 7
From the measurements of capacitance and tan, the complex capacitance is calculated. The complex capacitance and not complex relative permittivity is used because the geometric capacitance changes with degree of curing. ( T, C
) = C 0 ( ) ( (
= C0 ( )
T,
) i (
T,
(T ) + (
)) = (T ) + ( ) i
T,
0
T, )
(7.12)
The master curve shifting was also done only with the imaginary part of the complex capacitance under the assumption that conductivity could be neglected. If conductivity was visible it was subtracted, but caution must be taken for slopes close to -1 since it can be low frequency dispersion (LFD). To use the real part of the complex capacitance the strongly temperature dependent high-frequency component of the relative permittivity must be individually subtracted from each measured temperature level. In Figure 7.15, the shifts along the -axis and the amplitude axis are defined from shifting measured data at temperature T to the measured data at the reference temperature Tref. When finding the shifts along both axes, the reference point can actually be arbitrarily put on the graph whereas the measured curves should be shifted to one of the measured temperatures.
log(C)
C(, T, ) C(ref, Tref, )
ln(Xshift )
ln(Yshift )
T Tref ref Figure 7.15
log( )
With master curve shifting technique measured imaginary part of the complex capacitance is shifted both in -direction and in amplitude-direction from temperature T to reference temperature Tref.
The natural logarithmic shift along the -axis (Xshift) and along the C-axis (Yshift) can be written as
ln (X shift ) = ln (
ref
) ln
ref
Ex E 1 = x k BTref k B T A x (T, Tref ) 1 2 3

ref =const ref
ln (Yshift ) = ln (C(
, Tref , ) ln (A y (T, Tref ) C(
, Tref , ) =
Ey k BTref 1 2 3
=const
E 1 + y kB T
(7.13)
To find the activation energies Ex,y, which can be temperature dependent, the natural logarithmic shifts are plotted against 1/T. It is mostly common to talk about activation energies in the -axis direction but here also activation in amplitude-axis direction was introduced even though it might be more abstract to interpret. 148
Chapter 7
Experience shows that it is somewhat sensitive in which order the measured curves are shifted and for each measurement, several possibilities were tried. Final decision was always made by visually looking at a plot of the shifted curves. A numerical algorithm was written, minimising the relative error in a fixed frequency window in both -axis direction and in amplitude-axis direction, using a simplex search method [73].
7.3.1
The resin rich-mica tape with woven glass as carrier material
The temperature dependence, -50C<T<50C, for the resin-rich mica tape with woven glass as carrier material was studied at different degrees of curing, 0, 0.4, 0.6, 0.8, 0.9, 1. Each degree of curing was assumed to have Arrhenius activation, shown in equations (7.10) and (7.11). For each degree of curing the natural logarithmic shift in both -axis direction (Xshift) and in amplitude-axis direction (Yshift) was plotted as a function of 1/T and the final master curve was plotted at T=20C.
7.3.1.1
Degree of curing, 0
In Figure 7.16 (A), the natural logarithmic shifts in frequency (Xshift) and amplitude (Yshift) directions are plotted against 1/T for capacitance and tan measurements in the temperature range -50C to 50C.
2 EY1 0.08 eV 0 ) -2 -4 -6 -8 -10 0.003 EX1 = 1.26 eV E
X2
10-6 EY2 = 0.15 eV Complex capacitance [F] 10-7 T = 20C p 1.2 MPa
shift
and Y
10
-8
shift
C'() 10
-9
ln( X
= 0.67 eV
10-10 C''() 10
-11
-50C < T < 50C p 1.2 MPa 0.0035 1/T [K-1] 0.004 0.0045
10-4
10-2
100
102
104
106
Frequency [Hz]
Figure 7.16
(A) Natural logarithmic shifts in frequency and amplitude directions for the uncured resin-rich mica tape. (B) Master curve, complex capacitance, at T=20C for 0.
The master curve shifting was split up in two temperature intervals first, -50C to 0C and second 0C to 50C. This split up was done because of the clear change in activation energy (Xshift) around 0C, see Figure 6.5, and because measurements were done in different frequency intervals. The temperature dependence for the uncured system, 0, can be described in the range, -50C<T<-10C, by activation energies of Ex0.7 eV and Ey0.2 eV and in the range, -10C<T<50C, by activation energies of Ex1.3 eV and Ey0.1 eV.
149
Chapter 7
In Figure 7.16 (B), measured data in form of complex capacitance is plotted at T=20C, shifted according to calculated activation energies. There are some fluctuations in the master curve which illustrates that the uncured system is not thermally activated in a simple way. However, the overall shape of the measured curves in the temperature interval studied here is kept.
7.3.1.2
For the degree of curing, 0.4, the natural logarithmic shifts in frequency (Xshift) and amplitude (Yshift) directions to find the master curve are plotted in Figure 7.17 (A) for a temperature range of -50C to 50C. It is also here, like for 0, that the temperature dependence is split up in two regions. The first region is -50C<T<0C with activation energies Ex0.7 eV and Ey0.1 eV and the second region is 0C<T<50C with activation energies Ex1.3 eV and Ey0.1 eV. The activation energies here for 0.4 are the same as for 0 but the temperature at which the change occurs is shifted towards higher temperatures. This is also what is expected since the network formation in the epoxy-resin has started limiting the number of polar groups and their movements, needing higher temperatures and more energy, to start rotational movements.
10 E 5 ln( Xshift and Yshift )
X1
10 = 1.25 eV -50C < T < 50C p 1.2 MPa Complex capacitance [F]
-8
T = 20C p 1.2 MPa 10-9 C'()
E 0.08 eV
Y
10-10
C''()
-5
EX2 = 0.68 eV
10-11
-10 0.003
10 0.0035 1/T [K ]
-1
-12
0.004
0.0045
10
-6
10
-4
10
-2
10
10
10
10
10
10
10
Frequency [Hz]
Figure 7.17
(A) Natural logarithmic shifts in frequency and amplitude directions for the resin-rich mica tape cured to 0.4. (B) Master curve, complex capacitance, at T=20C for 0.4.
In Figure 7.17 (B), the master curve at T=20C from measured data in form of complex capacitance is plotted shifted according to calculated activation energies. There are still some fluctuations in the master curve as found for 0. Most of these come from the change in activation energy at around T=-10C, which also changed the shape of the measured complex capacitance curves, see Figure 6.11. This was also seen for 0 but the change came at T=0C and were a bit more difficult to see because measurements were done in a smaller frequency interval, see Figure 6.5. Comparing the overall shape of the master curves for 0 and 0.4 it is seen that the saddle point is shifted towards lower frequencies with increasing and there is a visible high frequency peak entering the measurement window.
150
Chapter 7
7.3.1.3
For the degree of curing 0.6 it can be seen in Figure 7.18 (A) that only one activation energy Ex0.7 eV in the frequency direction and one activation energy Ey0 eV in the amplitude direction are needed. This is for the temperature range -60C<T<50C and compared to 0 and 0.4 the activation energy is the same but the above seen transition is shifted out of the measurement window towards higher temperatures (lower frequencies). The thermal activation in the amplitude direction is reduced to zero leaving only the traditional shift in the frequency direction.
4 2 0 -2 -4 -6 -8 -10 0.003 10-12 10-4 EX = 0.68 eV -60C < T < 50C p 1.2 MPa Complex capacitance [F] E 0.03 eV
Y
10-8 T = 20C p 1.2 MPa 10-9 C'()
ln( Xshift and Yshift )
10
-10
C''() 10
-11
0.0035
0.004 1/T [K-1]
0.0045
0.005
10-2
100
102
104
106
108
1010
Frequency [Hz]
Figure 7.18
The master curve for 0.6 at T=20C is presented in Figure 7.18 (B). The master curve fit becomes better with increasing degree of curing, which shows that the network formation in the epoxy-resin creates a more rigid structure. However, there is still some spread in data for the lower frequencies. The high frequency peak is shifted towards lower frequencies compared to 0.4.
7.3.1.4
When reaching the degree of curing 0.8 it can be seen in Figure 7.19 (A) that only one activation energy is needed Ex0.6 eV in the frequency direction and none in the amplitude direction since Ey0 eV. The fit to one activation energy in the temperature range 50C<T<50C is good and compared to 0, 0.4 and 0.6 it can bee seen that the activation energy is decreasing with increasing degree of curing. For 0.8 the master curve at T=20C is presented in Figure 7.19 (B). Like also seen above, for 0, 0.4 and 0.6, the master curve fit becomes better with increasing degree of curing, which shows that the network formation continues in the epoxy-resin creating more cross-links and a more rigid structure. The deviation from one single curve seen in Figure 7.19 (B) for the lower frequencies is probably just a numerical artifact. However, the numerical routine works well especially since it can be difficult to find the master curve when measured data have small dispersion or no distinct features like peaks etc. The high 151
Chapter 7
frequency peak here for 0.8 is shifted towards lower frequencies compared to the lower degrees of curing especially 0.4 and 0.6.
4 -50C < T < 50C p 1.2 MPa 2 ln( Xshift and Yshift ) 0 -2 EX = 0.58 eV -4 -6 -8 0.003 10-12 10-4 E 0.02 eV
Y
10-8 T = 20C p 1.2 MPa Complex capacitance [F] 10-9 C'()
10
-10
C''() 10-11
0.0035 1/T [K ]
-1
0.004
0.0045
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.19
7.3.1.5
When increasing the degree of curing to 0.9 it can be seen in Figure 7.20 (A) that still only one activation energy in the frequency direction is needed Ex0.6 eV and Ey0 eV. The temperature range is also here -50C<T<50C. There is a tendency towards a small decrease in Ex but it is still within the uncertainty even though natural logarithmic shifts fit a straight line well. It is then clearer that the shifts in the amplitude direction are zero.
4 2 ln( Xshift and Yshift )
Y
10-8 -50C < T < 50C p 1.2 MPa E 0 eV Complex capacitance [F] 10-9 C'() T = 20C p 1.2 MPa
0 -2 EX = 0.56 eV -4 -6 -8 0.003
10
-10
C''() 10
-11
0.0035 1/T [K-1]
0.004
0.0045
10-12 10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.20
In Figure 7.20 (B) the master curve is plotted for the degree of curing 0.9 at T=20C. The measured data fit into one master curve very well confirming again that the master curve fit becomes better with increasing degree of curing. The high frequency peak is still 152
Chapter 7
shifted into the measurement window and there are some indications that maximum value is soon reached but it seems to be a rather wide peak.
7.3.1.6
Degree of curing, 1
The fully cured, 1, insulation was measured in a temperature interval -50C<T<170C starting with the lower temperatures. In the temperature region -50C<T<50C the activation energy decreases to Ex0.5 eV and Ey0 eV compared to lower degrees of curing. For the higher temperature region 50C<T<170C the activation energy increases to Ex1.3 eV and Ey0 eV but there is still only temperature activation in the frequency direction.
15 EX1 = 1.31 eV -50C < T < 170C p 1.2 MPa Complex capacitance [F] 10-9 C'() 10-8 T = 20C p 1.2 MPa
10 ln( Xshift and Yshift )
5 EY 0 eV 0
10
-10
C''()
10-11
-5
EX2 = 0.53 eV 10-12 10-8 10-6 10-4 10-2
-10 0.002
0.0025
0.003
0.0035
0.004
0.0045
100
102
104
106
108
1/T [K-1]
Frequency [Hz]
Figure 7.21
(A) Natural logarithmic shifts in frequency and amplitude directions for the fully cured resin-rich mica tape. (B) Master curve, complex capacitance, at T=20C for 1.
The shifted measured data fit well into a master curve at T=20C, see Figure 7.21 (B). The high frequency peak is visible but there are still no indications, like for 0.9, of a maximum value. In the low frequency region there is an increase in loss and a saddle point entering the measurement window. It is possible that this is the same saddle point as seen for 0 and 0.4 but shifted towards lower frequencies as degree of curing increases. Comparing the fully cured system in Figure 7.21 (B) with the uncured system in Figure 7.16 (B) it is clear that the fully cured system follows an Arrhenius activation, see equations (7.10) and (7.11), much better than the uncured system.
7.3.1.7
Comparison between different degrees of curing
In Figure 7.22 (A), the calculated activation energy in the frequency direction (Ex) is plotted against measured degree of curing . For 0 and 0.4 the weighted average is plotted (filled dots) since these two degrees of curing undergo a change in activation energy in the temperature interval studied, -50C<T<50C. An approximate linear relation can be found between activation energy and degree of curing, Ex1-0.46. From each degree of curing the master curve calculated from the complex capacitance at T=20C was taken and tan was calculated. Before plotted in Figure 7.22 (B) the set of data for each was least-square fitted to a sum of exponentials giving an average tan curve. 153
Chapter 7
1 -50C < T < 50C p 1.2 MPa 0.9 Activation energy [eV]

101 T = 20C p 1.2 MPa 100 =1 10
-1
0.8 E 1 - 0.46
X
tan()
=0 = 0.4 = 0.6
0.7 10-2 0.6 = 0.8 = 0.9
0.5 0 0.2 0.4 0.6 0.8 1 Degree of curing,
10-3 10-8 10-6 10-4 10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.22
(A) Relationship between calculated activation energy (Ex) in frequency direction and degree of curing. A filled dot indicates average activation energy in the frequency window studied. (B) Calculated tan master curve at T=20C for the different degrees of curing.
For lower <0.6 there is a shift of the saddle point towards lower frequencies but it is unclear what happens for higher . There is a high frequency peak, which is shifted towards lower frequencies with increasing , for 0.5< <1. A possible explanation could be that around 0.5, the final structure in the epoxy-resin network is reached and just more cross-links are added with increasing . The reduction in available cross-links, 1- , will be seen as a wide dipolar peak shifting towards lower frequencies. There is no clear activation, natural logarithmic shifts, in frequency or amplitude directions for the different degrees of curing but an attempt has been made in Figure 7.23 to shift the measured data in Figure 7.22 (B) to =0.
20 T = 20 C p 1.2 MPa 15 )
shift
10
T = 20 C p 1.2 MPa 10 E = 0.49 eV

1
and Y
10
10 tan( )
ln(X
10
-1
= = = = = =
0 0.4 0.6 0.8 0.9 1
shift
0 E -5 0 0.2 0.4 0.6 0.8 1 Degree of curing,

Y
10
0 eV
-2
10
-3
10
-4
10
-2
10
10
10
10
10
10
10
10
12
10
14
10
16
Frequency [Hz]
Figure 7.23
(A) Natural logarithmic shifts in frequency and amplitude directions for different degrees of curing. (B) Master curve, tan, at T=20C and at =0.
In Figure 7.23 (A), the natural logarithmic shifts in frequency and amplitude directions are plotted. The activation energies are calculated from the following relations, see also equation (7.13). 154
Modelling a resin-rich insulation system Ex Ex 1 ln(X shift ) = k T k T ref + B ref B 14 4 244 3

= const
Chapter 7
Ey 1 Ex ln(Yshift ) = + k T ref k T B B ref 14 4 244 3

= const
(7.14)
Activation energy in the frequency direction was calculated to Ex0.5 eV and in amplitude direction to Ey0 eV. The fit of the logarithmic shifts to a straight line is good. The tan master curve at T=20C and at =0 is plotted in Figure 7.23 (B). The fit to one single curve is not perfect but not very wrong. However, the physical meaning of the curve at the very high frequency end can be discussed.
7.3.1.8
Comparison between fully cured samples cured under Factory, Soft and Hard temperature paths
The temperature dependence was also studied for three samples that were cured under the different temperature paths Factory, Soft and Hard, see Figure 7.9 (A), but to the same degree of curing 0.9. Capacitance and tan were measured after each temperature path and the master curves have been calculated for these profiles in Figure 7.24, Figure 7.25 and Figure 7.26.
2 "Factory" E 0
shift Y
10
0 eV
-8
"Factory" C'( ) Complex capacitance [F ] 10

-9
T = 20 C p 1.2 MPa
) and Y -2 E = 0.60 eV
X
10
-10
shift
ln( X
-4
C''( ) 10
-11
-6 -50 C < T < 50 C p 1.2 MPa -8 0.003 10 0.0035 1/T [K ]

-1 -12
0.004
0.0045
10
-4
10
-2
10
10
10
10
10
Frequency [Hz]
Figure 7.24
(A) Shifts in frequency and amplitude directions for the resin-rich mica tape cured under a Factory like temperature path, 0.9. (B) Master curve, complex capacitance, at T=20C.
For the Factory temperature path, the natural logarithmic shifts in the frequency direction and amplitude direction are found to be Ex0.6 eV and Ey0 eV, see Figure 7.24 (A). Calculations for the Soft temperature path give Ex0.5 eV and Ey0 eV, see Figure 7.25 (A), and for the Hard temperature path Ex0.6 eV and Ey0 eV, see Figure 7.26 (A). The Factory and Hard samples show similar activation energies as for the samples cured to 0.8, see chapter 7.3.1.4 and 0.9, see chapter 7.3.1.5. There is a difference for the Soft sample, which shows lower activation energy in the frequency direction, more like the fully cured sample in chapter 7.3.1.6. However, master curve calculations can be sensitive too, for example, the order in which the measured curves are shifted and therefore before making 155
Chapter 7
too extensive conclusions more samples should be studied. Nevertheless, these master curve calculations show consistency with calculations made for the different degrees of curing earlier in this chapter although fewer measurements in the temperature range were made here.
2 "Soft" 0
shift
10 E
Y
-8
0 eV
"Soft" C'( ) Complex capacitance [F ] 10

-9
T = 20 C p 1.2 MPa
) and Y -2 E = 0.52 eV
X
10
-10
shift
ln( X
-4
C''( ) 10
-11
-6 -50 C < T < 50 C p 1.2 MPa -8 0.003 10 0.0035 1/T [K ]

-1 -12
0.004
0.0045
10
-4
10
-2
10
10
10
10
10
Frequency [Hz]
Figure 7.25
(A) Shifts in frequency and amplitude directions for the resin-rich mica tape cured under a Soft temperature path, 0.9. (B) Master curve, complex capacitance, at T=20C.
In Figure 7.24 (B), Figure 7.25 (B) and Figure 7.26 (B) it can be seen that the calculated master curves look very much the same. For the three samples tan were calculated and plotted on the same graph, see Figure 7.27, showing that it is not possible to separate the three different samples. It is not possible with a capacitance and tan measurement to say anything about the path (temperature path) the sample reached its degree of curing. However, the different degrees of curing can be separated.
2 "Hard" E 0 Complex capacitance [F ] 10
-9 Y
10
0 eV
-8
"Hard" C'( )
T = 20 C p 1.2 MPa
) and Y
shift
-2 E = 0.59 eV -4
X
10
-10
shift
ln( X
-6
C''( ) 10
-11
-8
-50 C < T < 50 C p 1.2 MPa 10 0.0035 1/T [K ]

-1 -12
-10 0.003
0.004
0.0045
10
-4
10
-2
10
10
10
10
10
Frequency [Hz]
Figure 7.26
(A) Shifts in frequency and amplitude directions for the resin-rich mica tape cured under a Hard temperature path, 0.9. (B) Master curve, complex capacitance, at T=20C.
156

10
-1
Chapter 7
T = 20 C p 1.2 MPa
"Factory" "S oft" "Hard"
tan( )
10
-2
10
-3
10
-4
10
-2
10
10
10
10
10
Frequency [Hz]
Figure 7.27
Comparison between measured tan for the resin-rich mica tape samples cured under Factory, Soft and Hard temperature paths. All samples reached a degree of curing 0.9.
7.3.2
The resin rich-mica tape with polyester film (PET) as carrier material
The temperature dependence, 10C<T<50C, for the resin-rich mica tape with polyester film (PET) as carrier material was studied at different degrees of curing, 0, 0.1, 0.3, 0.5, 0.8, 1. Each degree of curing was assumed to have Arrhenius activation as shown in equations (7.10) and (7.11). For each degree of curing the natural logarithmic shift in both the -axis direction (X shift) and in the amplitude-axis direction (Y shift) was plotted as a function of 1/T and the total master curve at T=20C.
7.3.2.1
Degree of curing, 0
For the uncured resin-rich mica tape with polyester film (PET) the natural logarithmic shifts in frequency (Xshift) and amplitude (Yshift) directions are plotted against 1/T for capacitance and tan measurements in the temperature range 10C to 50C, see Figure 7.28 (A). The temperature dependence for the uncured system, 0, can be described with activation energies Ex1.6 eV and Ey0.1 eV. In Figure 7.28 (B), measured data in form of complex capacitance is plotted as a master curve at T=20C, shifted according to calculated activation energies. There are some fluctuations in the master curve which illustrates that the uncured system is not thermally activated in a simple way. In the loss part there are two peaks compared to the resin-rich mica tape with woven glass, which has only a saddle point, see Figure 7.16. This can be explained by the solid polyester film, which will create a barrier effect due to moving polar molecules in the epoxy-resin. The polyester film (PET) itself is rather polar and will have a loss peak that could also be seen in this measurement window and temperature. However, it is also for this system difficult to distinguish between effects coming from each material and effects coming from geometrical combinations of the materials.
157
Chapter 7
2 10C < T < 50C p 1.2 MPa Complex capacitance [F] 0 )
shift

10-8 T = 20C p 1.2 MPa 10-9 C'()
EY = 0.08 eV -2
and Y
C''() 10
-10
shift
ln( X
-4 EX = 1.56 eV -6
10
-11
-8 0.003
0.0032 1/T [K-1]
0.0034
0.0036
10-12 10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.28
(A) Shifts in frequency and amplitude directions for the uncured resin-rich mica tape with polyester film (PET) as carrier material. (B) Master curve, complex capacitance, at T=20C for 0.
7.3.2.2
The resin-rich mica tape with polyester film (PET) was then cured to 0.1 and the shifts in frequency (Xshift) and amplitude (Yshift) directions are plotted against 1/T for capacitance and tan measurements in the temperature range 10C to 50C, see Figure 7.29 (A). The temperature dependence for the system cured to 0.1 can be described with activation energies Ex2.0 eV and Ey0 eV. The activation energy in the frequency direction, Ex, increased compared to 0 which is in contradiction to the behaviour seen for the resin-rich mica tape with woven glass.
4 2 0 E 0 eV
Y
10C < T < 50C p 1.2 MPa Complex capacitance [F]
10-8 C'() T = 20C p 1.2 MPa
10-9
and Y
shift
-2 -4 -6 -8 -10 0.003 EX = 1.98 eV
10
-10
C''()
ln( X
shift
10
-11
0.0032 1/T [K-1]
0.0034
0.0036
10-12 10-5
10-3
10-1
101
103
105
107
Frequency [Hz]
Figure 7.29
(A) Shifts in frequency and amplitude directions for the resin-rich mica tape with polyester film (PET) as carrier material cured to 0.1. (B) Master curve, complex capacitance, at T=20C for 0.1.
The master curve at T=20C is plotted in Figure 7.29 (B), shifted according to calculated activation energies. The two loss peaks are still visible in the measurement window but they
158
Chapter 7
are shifted towards lower frequencies compared to 0. Measured data, 0.1, fit the calculated master curve well.
7.3.2.3
The shifts in frequency (Xshift) and amplitude (Yshift) directions are plotted against 1/T for measurements in the temperature range 10C to 50C, see Figure 7.30 (A) for a sample of resin-rich mica tape with polyester film (PET) which was cured to 0.3. Calculated activation energies for 0.3 are Ex2.2 eV and Ey0 eV. The activation energy in the frequency direction, Ex, is still increasing compared to 0 and 0.1 which is a behaviour not seen for the resin-rich mica tape with woven glass.
2 EY 0 eV 0 -2 10C < T < 50C p 1.2 MPa -4 EX = 2.17 eV -6 -8 -10 0.003 10-12 10-6 Complex capacitance [F] 10-9 ) C'() 10-8 T = 20C p 1.2 MPa
and Y
shift
10
-10
shift
ln( X
C''() 10
-11
0.0032 1/T [K ]
-1
0.0034
0.0036
10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.30
In Figure 7.30 (B) the master curve at T=20C is plotted, shifted according to calculated activation energies. For this degree of curing there is only one loss peak visible, compared to for 0 and 0.1. The loss peak continuous to be shifted towards lower frequencies as the degree of curing increases. Measured data, 0.3, fit well to a single master curve.
7.3.2.4
The degree of curing is increased to 0.5 for the resin-rich mica tape with polyester film (PET) and shifts in frequency (Xshift) and amplitude (Yshift) directions are plotted in Figure 7.31 for measurements in the temperature range 10C to 50C. Activation energies are found to be Ex1.0 eV and Ey0.2 eV. The activation energy in the frequency direction, Ex, decreased compared to 0.3 whereas activation energy in the amplitude direction, Ey, increased. Also for this degree of curing, 0.5, a behaviour is seen which is not the same as for the resin-rich mica tape with woven glass. The master curve at T=20C for the degree of curing 0.5 is plotted in Figure 7.31 (B). The only loss peak is now almost shifted out of the measurement window compared with 0.3. There is also a high frequency loss peak entering the measurement window. There is some spread in the master curve data, which could come from problems with the quality of 159
Chapter 7
the measurement data see Figure 6.68, especially since the activation energy in amplitude direction, Ey, increased so much.
1 0 -1 E = 0.15 eV
Y
10C < T < 50C p 1.2 MPa Complex capacitance [F]
10-8 T = 20C p 1.2 MPa 10-9 C'()
and Y
shift
-2 -3 -4 E = 1.03 eV
X
10
-10
shift
ln( X
C''() 10
-11
-5 -6 0.003 10-12 10-4
0.0032 1/T [K ]
-1
0.0034
0.0036
10-2
100
102
104
106
Frequency [Hz]
Figure 7.31
7.3.2.5
The resin-rich mica tape with polyester film (PET) is here cured to 0.8 and in Figure 7.32 (A), shifts in frequency (Xshift) and amplitude (Yshift) directions are plotted for measurements in the temperature range 10C to 50C. Activation energies are found to be Ex0.7 eV and Ey0 eV. The activation energy in the frequency direction, Ex, continuous to decrease compared to 0.5. For the activation energy in the amplitude direction, Ey, it is now back to zero like for 0.1 and 0.3.
3 2.5 2
shift
10-8 10C < T < 50C p 1.2 MPa Complex capacitance [F] 10-9 C'() T = 20C p 1.2 MPa
) and Y 1.5 1 0.5 0 -0.5 -1 0.003 EY 0.02 eV
E = 0.74 eV
X
10
-10
ln( X
shift
10
-11
C''()
0.0032 1/T [K-1]
0.0034
0.0036
10-12 10-4
10-2
100
102
104
106
Frequency [Hz]
Figure 7.32
The fit to a master curve at T=20C for the measured data is good, see Figure 7.32 (B). The loss peak seen for lower degrees of curing is now totally shifted out of the measurement 160
Chapter 7
window. The high frequency peak, which was first seen for 0.5 is here slightly shifted toward lower frequencies.
7.3.2.6
Degree of curing, 1
For the fully cured resin-rich mica tape with polyester film (PET) the shifts in frequency (Xshift) and amplitude (Yshift) directions for measurements in the temperature range -50C to 50C can be seen in Figure 7.33 (A). Activation energies are found to be Ex0.6 eV and Ey0 eV. The activation energy in the frequency direction, Ex, continuous to decrease compared to 0.8 and the activation energy in the amplitude direction, Ey, is still zero like for 0.1, 0.3 and 0.8.
2 EY 0 eV 0 -2 -50C < T < 50C p 1.2 MPa -4 E = 0.62 eV
X
10-8 T = 20C p 1.2 MPa Complex capacitance [F] 10-9 C'()
and Y
shift
10
-10
ln( X
shift
-6 -8 -10 0.003
10
-11
C''()
0.0035 1/T [K ]
-1
0.004
0.0045
10-12 10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.33
(A) Shifts in frequency and amplitude directions for the fully cured resin-rich mica tape with polyester film (PET) as carrier material. (B) Master curve, complex capacitance, at T=20C for 1.
The master curve at T=20C for the fully cured sample is plotted in Figure 7.33 (B). There is some spread in measured data especially in the loss part for the higher frequencies. Still the high frequency peak can be seen and it has the same position as for 0.8. The low frequency part of the master curve for the fully cured sample is almost flat compared to the lower degrees of curing.
7.3.2.7
Comparison between different degrees of curing
Calculated activation energy in the frequency direction (Ex) is plotted against measured degree of curing in Figure 7.34 (A). There is an increase in activation energy for the lower degrees of curing, followed by a drop in activation energy around 0.4 and then the activation energy is decreasing with increasing degree of curing. No simple relationship between activation energy and degree of curing can be found as in the case with the resin-rich mica tape with woven glass see Figure 7.22 (A). In Figure 7.34 (B), tan was calculated from each master curve at T=20C for the different degrees of curing studied. There is a clear shift in frequency direction for the lower degrees of curing, 0<<0.5, where the two peaks are gradually shifted towards lower frequencies with increasing degree of curing. For the higher degrees of curing, it is difficult to see a shift 161
Chapter 7
mainly because the dispersion is smaller. However, a high frequency peak is shifted into the measurement window with increasing degree of curing.
2.5 10C < T < 50C p 1.2 MPa Activation energy [eV] 2 10-1 tan() 1.5 = 0.5 10-2 1 = 0.8 =1 =0 = 0.1 = 0.3 100 T = 20C p 1.2 MPa
0.5 0 0.2 0.4 0.6 0.8 1 Degree of curing,
10-3 10-6
10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.34
(A) Relationship between calculated activation energy (Ex) in frequency direction and degree of curing. (B) Calculated tan master curve at T=20C for the different degrees of curing.
Master curve shifting was also done for the different degrees of curing in Figure 7.34 (B) and the result from this can be seen in Figure 7.35. The activation energies were calculated to be Ex0.4 eV in the frequency direction and to Ey0 eV in the amplitude direction using the same definition as used for the resin-rich mica tape with woven glass see equation (7.14). The fit of the natural logarithmic shifts to a straight line in Figure 7.35 (A) is good. The resulting master curve in tan at T=20C and at =0 is plotted in Figure 7.35 (B). The fit to one single master curve is good for 00.5 whereas for 0.5<1 there is some deviation especially in the low frequency end. The higher degrees of curing are shifted to very high frequencies in order to fit the master curve and the physical meaning of such a high frequency peak, which is revealed here, is not fully understood.
20 T = 20 C p 1.2 MPa 15
shift
10
= = = =
0 0.1 0.3 0.5
T = 20 C p 1.2 MPa
10 10 E = 0.36 eV
X
-1
and Y
shift
ln(X
5 10 0 E -5 0 0.2 0.4 0.6 0.8 1 Degree of curing,

Y -2
tan( )
= 0.8 = 1 0 eV
10
-3
10
-4
10
-2
10
10
10
10
10
10
10
10
12
10
14
Frequency [Hz]
Figure 7.35
(A) Natural logarithmic shifts in frequency and amplitude directions for different degrees of curing. (B) Master curve, tan, at T=20C and at =0.
162
Chapter 7
7.4
Modelling of the dielectric response from resin-rich mica tapes
Two different insulation systems have been studied here, both are resin-rich mica tape based systems but they have different tape carriers such as woven glass and polyester film (PET). Dielectric measurements have been made in the frequency domain as a function of temperature and degree of curing. The measured complex relative permittivity for this type of insulation system can be expressed as
= ( T, 7W ) - i ( T, (T(t) ))
(7.15)
where is the angular frequency, T is the temperature and (T(t)) is the degree of curing at time t after following the temperature path T=T(t). As seen so far the temperature can be successfully modelled with Arrhenius activation especially for the higher degrees of curing. The change in degree of curing is more complex especially for the resin-rich mica tape with woven glass whereas the tape with polyester film can be shifted well along the frequency axis with change in degree of curing. A simple engineering model will be discussed here to describe the dielectric response at fixed degree of curing and temperature for the resin-rich mica tapes studied. The model is based on that the tapes have a macroscopic distribution of materials that is not random.
7.4.1
Network model
The idea is to build up the dielectric response from the resin-rich mica tapes from the dielectric responses of the materials in the tapes, mica, epoxy-resin and woven glass or polyester (PET) depending on tape carrier material. There is a macroscopic structure in each tape and by putting the materials in a similar combination a total dielectric response can be reached that is close to what is measured. The simplest combination of materials is series and parallel combinations but to extend the possible combinations a network according to Figure 7.36 has been studied. It is an nm network with both vertical and horizontal impedances all with dispersion.
n u11 m eV11 ZV11 u21 eV21 ZH11 eV12 u12 ZV12 u22 ZH12 eV1n eV2n u1n ZV1n u2n ZV2n U um-11 eVm-11 ZHm-21 um-12 ZHm-22 eHm-22 ZVm-12 um2 eVm-1n um-1n ZVm-1n umn E
H V ZV21 e 11 e 22
H ZV22 e 12
eHm-21 V ZVm-11 e m-12 um1
Figure 7.36
Generalized network model to represent the dielectric response from the resinrich mica tapes. The impedances are mica, epoxy-resin and woven glass or polyester (PET) depending on which tape is studied. 163
Chapter 7
This network can be solved in both time and which is done here in the frequency domains. The time domain solution, see Appendix D, is more general but also more time consuming to solve and for this discussion the frequency domain solution is enough [93]. To solve the network in Figure 7.36 in the frequency domain, the first step is to define potentials ux,y, voltage drops, both vertical (V) eVx,y and horizontal (H) eHx,y and impedances, both vertical (V) ZVx,y and horizontal (H) ZHx,y. The vertical and horizontal voltage drops over the complex impedances can be written as
eV x, y = u x, y u x +1, y eH x, y = u x +1, y u x +1, y +1
x = 1..m - 1, y = 1..n x = 1..m - 2, y = 1..n - 1
(7.16)
where n and m defines the size of the network. The external voltage, E, with phase set to zero is applied over the network setting boundary conditions. This results in the following extra equations
E = u 1, y u m, y y = 1..n
(7.17)
The equations (7.16) and (7.17) can be written in matrix from as e = b A0 u (7.18)
where e is a vector with potential drops, b a vector with external voltage sources, A0 a matrix describing the nodes in the network and u a vector with the potentials in each node of the network. The vector b can actually contain several external voltage sources. The matrix A0 will be sparse and divided into two parts representing the vertical and horizontal nodes. The next step is to use Ohms law to define the currents through all the complex impedances. The vertical and horizontal currents can be written as
i i
V x, y
= =
eV x, y ZV x, y H e x, y ZH x, y
x = 1..m - 1, y = 1..n
(7.19)
x = 1..m - 2, y = 1..n - 1
H x, y
It is here that the frequency domain solution has its big advantages since the current can be written as a division between voltage drop and corresponding impedance while in the time domain that relation is more complex (convolution integral) see equations (3.21) and (3.27). The relation between current and voltage drop can be written in matrix form as
i = Y e
(7.20)
where in this case Y is a matrix with only elements on the diagonal, which makes it easy to calculate Y-1. The next step is to use Kirchoffs current law, summing up the currents in each node to zero. Here is also the second boundary condition entered, no currents are allowed to enter or leave the network from the sides. These conditions give the following equations
164

V H i V x = 1..m - 2 x,1 i x +1,1 i x,1 = 0 V V H H i x,y+1 i x +1,y+1 + i x,y i x,y+1 = 0 x = 1..m - 2, y = 1..n - 2 i V i V + i H = 0 x = 1..m - 2 x,n x +1,n x,n-1
Chapter 7
(7.21)
In addition, these equations can be written in matrix form as

AT 0 i = 0
(7.22)
where A0 is the matrix describing the nodes in the network. Looking closer at A0 it will turn out that each row can be summed up to zero, which proves that A0 has linearly dependent columns. This means that more information is needed to be able to solve this problem. The solution is to fix one point in the network to a known potential, which is most commonly done by grounding that point. The lower side of the applied voltage, E, is in this case grounded. This is written as
u m, y = 0 y = 1..n A 0 A
(7.23)
The columns in A are now linearly independent and the following equation system can be constructed
Y 1 i = Y e = Y (b A u ) T T A A i = 0 A i b = 0 u 0
(7.24)
This equation system can be straightforwardly solved and the solution will give all currents and potentials in the network. The total current through the network can be written as
V V V V i tot = i1,1 + .. + i 1, n = i m 1,1 + .. + i m 1, n Z tot =
E i tot
(7.25)
Knowing the total current the total impedance of the network can be determined and from that the total complex pemittivity as follows
tot (
) i
tot (
)=
1 tot i C0 Z tot
(7.26)
However, to determine the total complex relative permittivity the total geometric capacitance C0 is needed. To calculate C0 the equation system in (7.24) must first be solved with only air inside (r=1) and then once again with all the materials. Therefore, by entering the vertical and horizontal impedances, geometry and complex pemittivity, the total resulting complex relative permittivity of the network can be determined.
7.4.2
Dielectric response of the resin-rich mica tape with woven glass as carrier material
The carrier material in this tape is woven glass and on one of its sides, a mixture of epoxyresin and mica is applied. This mixture consists of epoxy-resin with mica flakes where the flakes are randomly positioned but parallel to the plane of the tape see Figure 7.37.
165
Chapter 7
Figure 7.37
Side view from a cross-section of one layer of resin-rich mica tape with woven glass carrier. Woven glass at the top and the epoxy-mica mixture below.
The total dielectric response of this composite reflects both the constituent materials and the geometrical structure. Most of the dielectric response originates from the epoxy-resin part. The mica and woven glass will be seen as a barrier to charge movements. Especially in the uncured state surface conduction and polarisation play important roles. In the beginning of the curing, the rate of reaction is high especially for high temperatures and as curing proceeds, the rate of reaction decays which is also reflected in the dielectric response. In Figure 7.38, two simple network models of the epoxy-mica tape are illustrated. These models take into account the geometrical structure of the tape and should be seen as an engineering alternative to formulas describing chaotic mixtures see chapter 3.9. Complex impedances of the constituent components scaled with their volume fraction are used as circuit elements. The S-model is a one-dimensional model only taking in account effects perpendicular to the epoxy-mica tape. The H-model is an extension of the S-model allowing also in-plane conduction and polarisation.
H-model S-model ZWoven glass ZEpoxy ZEpoxy ZEpoxy ZEpoxy ZMica ZMica
ZWoven glass
Figure 7.38
Two simple network models for the resin-rich mica tape system with woven glass as carrier material.
The epoxy-mica tape is built up from three main materials woven glass, epoxy-resin and mica see Figure 7.37 and Figure 6.3. Since it was not possible to get exactly these three materials, similar ones were chosen for the modelling. The dielectric response for woven glass and mica were measured whereas typical responses for uncured and fully cured epoxy-resin were found in literature. In Figure 7.39, measured temperature dependence and dielectric response of muscovite mica are shown together with data found in literature [40]. This data was used in the H- and S-models to represent the frequency dependence of mica. It was assumed that the dielectric response of mica was independent of the epoxy-mica tapes degree of curing. In Figure 7.40, measured temperature dependence and dielectric response are shown of a thick tape of woven glass. The woven glass was so thick that the amount of air affecting the real part of the complex relative permittivity was neglected along with the effect of the thin silicone coating, which was on the woven glass. In addition, the assumption was made that the dielectric response of the woven glass did not change with the epoxy-mica tapes degree of curing.
166

12 Measured 10 8 EX = 0.75 eV 6 4 2 EY 0 eV 0 -2 0.002 10C < T < 190C p 6 kPa Complex pemittivity,* 102 '() 10
1
Chapter 7
T = 25C
and Y
shift
100 10 10 10
-1
Measured Literature ''()
ln( X
shift
-2
-3
0.0025
0.003 1/T [K-1]
0.0035
0.004
10-4 10-8 10-6 10-4 10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.39
12
Dielectric response of muscovite mica measured and found in literature [40]. (A) Temperature dependence and (B) Complex relative permittivity.
103 Measured 10C < T < 190C p 6 kPa Complex pemittivity,* T = 25C 10
2
10 8
shift
and Y
EX = 0.72 eV 6 4 2 EY 0 eV 0 -2 0.002
101 10
0
'()
shift
ln( X
Measured 10-1 10
-2
''()
0.0025
0.003 1/T [K ]
-1
0.0035
0.004
10-3 10-8
10-6
10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.40
Measured dielectric response of a thick woven glass tape. (A) Temperature dependence and (B) Complex relative permittivity.
The third component in the resin-rich mica tape with woven glass is the epoxy-resin, which is here, modelled in two different ways. In chapter 7.4.2.1, the epoxy-resin will be modelled with conductivity and a Cole-Cole expression and in chapter 7.4.2.2 it will be modelled with conductivity and a LFD expression. The uncured and fully cured state of the tape will be modelled but in principal, any degree of curing could be studied.
7.4.2.1
Assumption epoxy-resin: dipole + conductivity
The first assumption is based on what is found in literature about the dielectric behaviour of epoxy-resins during curing. The dielectric response can be divided into a conductive part mostly due to ionic impurities and a dipolar part originating from rotational diffusion of polar molecules. The total complex relative permittivity of the epoxy-resin with the dipolar part modelled with the Cole-Cole expression can be written as [71], [107]
167
Chapter 7
Epoxy (
)=
1 + (i
)b
(7.27)
0
where is conductivity, high-frequency component of the relative permittivity, s static permittivity, centre time for the dipole relaxation peek and b is the width of the peak. When modelling the uncured resin-rich mica tape it is important to remember that in the tape there is a B-stage epoxy-resin with a small volatile content. This will effect the parameters in equation (7.27) especially conductivity but also the position and width of the dipolar peak compared to what is found in literature for ordinary uncured epoxy-resin mixtures. Parameters used in equation (7.27) are the following =3 s = 30 Uncured epoxy, = 0 : = 10 2 (s ) b = 0.4 10 1 m -1 10
(7.28)
) (measured)
Conductivity was measured and the rest of the parameters were all estimated so that the total calculated response from the H-model in Figure 7.38 fit to measured data, with mica see Figure 7.39 and woven glass see Figure 7.40. The results from putting together the materials according to the H- and S-models are shown in Figure 7.41. The effect is also shown from changing the geometry of the parallel impedance in the H-model to 10% and 1% of the epoxy impedance, which is the same as gradually going from the H- to the S-model.
102 H-Model C =0.1*C
0 // 0 Epoxy
10
T = 20C p 1.2 MPa =0
C =0.01*C
0 // 0
0 Epoxy
tan()
10
Measured data 10-1 S-Model 10-2 10-4 10-2 100 102 104 106
Frequency [Hz]
Figure 7.41
Comparison between measured and modelled data for uncured, =0, epoxymica tape with woven glass as carrier material. The epoxy-resin is modelled with a conductive part and a dipolar part.
It is clearly seen that in the uncured case it is necessary to allow horizontal charge movement, H-model, around the woven glass and mica to reach a similar dielectric response as measured. The S-model shows a strong barrier effect and it is not possible with that model and assumed dielectric behaviour of the epoxy-resin, see equation (7.27), to get a response close to what is measured. 168
Chapter 7
For the fully cured resin-rich mica tape, parameters for the epoxy-resin see equation (7.27) were chosen to similar values as found in literature. =3 =9 s Fully cured : = 1010 (s ) epoxy, = 1 b = 0.1 17 1 1 10 ( m ) (measured ) 3 6.5 s (7.29) Literature : 4 1010 (s ) b 0.35 15 1 1 2.5 10 ( m )
The basic idea here is that when curing proceeds the conduction will be reduced and the dipolar peak will be broaden as it moves towards lower frequencies. In Figure 7.42, the results are shown for the fully cured tape using both the H- and S-models with the parameters in equation (7.29) for the epoxy-resin model and mica see Figure 7.39 and woven glass see Figure 7.40. Here both models are showing the same behaviour but they are slightly shifted in frequency compared to measured data. There is still a difference between the H- and S-models but this difference is reduced compared to the uncured case. The large contribution from DC conduction in the bulk together with LFD originating from the glass and mica surfaces as seen for the uncured tape is not seen here. All the materials in the fully cured tape have dielectric responses in the same magnitude resulting in that the geometrical configuration is less crucial compared to the uncured case. There is a high frequency peak seen in the measured data but not in the calculated and it is believed that this peak originates from a dipolar peak in the epoxy that is not described in equation (7.27).
100 H-Model 10-1 tan() T = 20C p 1.2 MPa =1
S-Model 10-2 Measured data 10-3 10-8
10-6
10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.42
Comparison between measured and modelled data for the fully cured, =1, epoxy-mica tape with woven glass as carrier material. The epoxy-resin is modelled with a conductive part and a dipolar part.
7.4.2.2
Assumption epoxy-resin: LFD + conductivity
The second epoxy-resin model is based on the assumption that in the uncured case the measured dielectric response from the resin-rich mica tape is mainly originating from the epoxy-resin. Looking at the measured data and especially at the slopes, the dielectric response from the epoxy-resin can be modelled as a sum of a conductive part and a low frequency dispersion (LFD) part. This can be written as 169
Chapter 7
Epoxy (
)=
+ A(i
) n +
(7.30)
0
where is conductivity, high-frequency component of the relative permittivity, A amplitude factor and n the slope of the LFD part. These parameters are estimated, except conductivity that is measured, in such away that the H-model in Figure 7.38 fit the measured data, see Figure 7.43. The parameters for the uncured resin-rich mica tape are found to be =3 A = 50 Uncured epoxy, = 0 : n = 0.3 10 10 1 m 1
(7.31)
) (measured)
In Figure 7.43, it can be seen that this model of the epoxy-resin fit measured data better than the epoxy-resin model used in Figure 7.41. This conclusion is based on the observation that the shape of the calculated curve is more like the measured curve, especially in the lower frequency region. The change of dielectric response going from the H-model via reduction of the parallel impedance (10% and 1%) in the H-model reaching the S-model with a clear barrier effect is observed in the same way as in Figure 7.41.
102 H-Model 10
1
C0 //=0.1*C0 Epoxy C =0.01*C

0 // 0 Epoxy
T = 20C p 1.2 MPa =0
tan()
10
Measured data 10-1 S-Model 10-2 10-4 10-2 100 102 104 106
Frequency [Hz]
Figure 7.43
Comparison between measured and modelled data for uncured, =0, epoxymica tape with woven glass as carrier material. The epoxy-resin is modelled with a conductive part and a LFD part.
The parameters in equation (7.30) for the fully cured resin-rich mica tape are chosen to be the same as corresponding parameters in equation (7.29). The amplitude factor A is then chosen so that the calculated dielectric response from the H-model has the same magnitude as the measured data. The parameters for the fully cured epoxy-resin are therefore chosen as =3 A = 5 10-1 Fully cured epoxy, = 1 : n = 0.1 17 1 1 10 ( m ) (measured )
(7.32)
170
Chapter 7
The results from the H- and S-models in Figure 7.44 are similar to what is found in Figure 7.42. This indicates that the two different models for the epoxy-resin seem to be equally good modelling the fully cured resin-rich mica tape. However, there is still a frequency shift between calculated and measured data and the high frequency peak seen in the measured data can not be calculated with made assumptions.
100 H-Model 10-1 tan() T = 20C p 1.2 MPa =1
S-Model 10-2 Measured data 10-3 10-8
10-6
10-4
10-2
100
102
104
106
108
Frequency [Hz]
Figure 7.44
Comparison between measured and modelled data for the fully cured, =1, epoxy-mica tape with woven glass as carrier material. The epoxy-resin is modelled with a conductive part and a LFD part.
7.4.3
Dielectric response of the resin-rich mica tape with PETfilm as carrier material
This tape is built up in the same way as the woven glass tape modelled above in chapter 7.4.2. It has a carrier material that is a polyester film (PET) and on one of its sides, there is a mixture of epoxy-resin and mica applied. This mixture consists of epoxy-resin with mica flakes which are also here assumed to be randomly positioned but parallel to the plane of the tape see Figure 7.45.
Figure 7.45
Cross-section of one layer of resin-rich mica tape with polyester film (PET) as carrier material. PET at the top and the epoxy-mica mixture below.
The H- and S-models used to represent the resin-rich mica tape with polyester film as tape carrier are shown in Figure 7.46. It is assumed here that the PET film will not allow the epoxy-resin to move freely as in the case of the woven glass carrier resulting in a barrier effect. This barrier effect is represented in both the H- and S-models by the ZPET impedance. However, the difference between the H- and S-models in this case is that the H-model allows free epoxy-resin movements around the mica flakes, which is not the case in the S-model (also a barrier).
171
Chapter 7
H-model

S-model ZPET ZEpoxy ZPET ZEpoxy ZEpoxy ZMica ZMica
ZPET
Figure 7.46
Two simple network models for the epoxy-mica tape system with polyester film (PET) as carrier material.
The dielectric response from the polyester film (PET, PolyEthylene Terephthalate) was found in literature [34]. It was chosen here to use the beta-relaxation peak (Cole-Cole expression) together with conductivity [51]. The total dielectric response in the temperature range -180C to 80C could therefore be written as 3.2 + 2 10 -4 (T - T0 ) s 3.7 - 1 10 -3 (T - T0 ) 1 1 1 ln 9.3 - 7.5 T - T 0 (7.33) b 0.3 + 2 10 -3 (T - T0 ) T0 236.7 (K ) -14 1 m 1 10
Literature : PET ( T ) =
) 1 + (i 14243
b relaxation peak
The same dielectric response as used for mica in the tape with woven glass is also used here for the tape with polyester film, see Figure 7.39. In addition, the epoxy-resin is also represented in the same way as for the tape with woven glass, see equation (7.27). The parameters for the uncured B-stage epoxy-resin are chosen to be =3 s = 15 Uncured epoxy, = 0 : = 10 1 (s ) b = 0.4 11 1 m 1 10
(7.34)
These parameters are chosen in the same range as for the uncured epoxy-resin in equation (7.28) and then optimised so that calculated dielectric response from the H-model fit measured data see Figure 7.47. The amount of each material, the geometrical capacitance, is calculated from the specification given by the manufacturer regarding tape composition see chapter 6.5 [55]. What is seen in Figure 7.47 is that the first peak probably originates from the dipolar peak in the epoxy-resin and the second peak, lower in frequency, originates from the barrier effect from the PET film. There is a small difference between the H-model and the Smodel but both models show similar behaviour. However, there is an indication that the allowed epoxy-resin movements around the mica flakes in the H-model fit measured data 172
Chapter 7
better. In the S-model for the lower frequencies, an additional peak is seen probably arising from barrier effects due to the mica flakes.
100 H-Model T = 20C p 1.2 MPa =0
10-1 tan() Measured data S-Model 10-2
10-3 10-5
10-3
10-1
101
103
105
107
Frequency [Hz]
Figure 7.47
Comparison between measured and modelled data for uncured, =0, epoxymica tape with polyester film (PET) as carrier material.
For the fully cured epoxy-resin, the same parameters are chosen as used in equation (7.29), which were similar to what is found in literature. =3 3 =9 6.5 s s Fully cured epoxy, = 1 : = 1010 (s ) Literature : 4 1010 (s ) b 0.35 b = 0.1 17 1 1 15 1 1 10 ( m ) 2.5 10 ( m )
(7.35)
Using these parameters and the dielectric responses from PET and mica, the H- and S-models were calculated and the results are shown in Figure 7.48. The H- and S-models give almost the same results showing like for the resin-rich mica tape with woven glass that the epoxyresin movements for the fully cured system can be neglected.
100 H-Model 10-1 tan() S-Model 10-2 Measured data 10-3 10-9 T = 20C p 1.2 MPa =1
10-7
10-5
10-3
10-1
101
103
105
107
Frequency [Hz]
Figure 7.48
Comparison between measured and modelled data for the fully cured, =1, epoxy-mica tape with polyester film (PET) as carrier material. 173
Chapter 7
The calculated dielectric responses show similar behaviour as the measured data. However, there is also here a shift in frequency between calculated and measured data as seen also in Figure 7.42. The high frequency peak seen in measured data is not detected in the calculated data. It is believed that this high frequency peak originates from the fully cured epoxy-resin. To be able to calculate this peak the model for the epoxy-resin in equation (7.27) must be modified with for example another Cole-Cole expression (dipolar peak). This additional peak is believed to behave in the same way as the first peak, moving towards lower frequencies as curing proceeds.
7.5
Summary
An Arrhenius activated reaction model was used in this chapter to model the epoxy-resin part in the two resin-rich mica tapes with woven glass and polyester film (PET) as carrier material. The parameters in the model were different for the two tapes and were determined from differential scanning calorimetry (DSC) measurements on uncured pieces. Correlation between measured and calculated degrees of curing was good for both tapes. With a reaction model for the epoxy-resin part, it was possible to calculate how the degree of curing changed for an arbitrary temperature path. Measured capacitance and tan as function of time during a temperature path could with the help of the reaction model be re-calculated as a function of degree of curing. It was found that an observed minima in the measured tan at f=50 Hz for the Factory temperature path correlated to a degree of curing of 0.8. This minimum in tan was found both for the laboratory samples (parallel-plate) and for the factory pressed stator bar. The dielectric response has been measured for both types of resinrich mica tapes at room temperature for laboratory samples cured to different degrees. Looking at the curve for 0.8 and at f=50 Hz a tan value will be given at room temperature. This means that when this minima is reached in tan at f=50 Hz during pressing under a Factory temperature path a tan value at room temperature can be estimated. Such estimation like this can be useful as a help to process the insulation in the press so that the insulation for sure will reach a certain tan value at room temperature outside the press. Measured values of capacitance and tan after pressing at room temperature are used to ensure good quality. However, caution must be taken when calculating the degree of curing when there is a temperature gradient in the sample, which was the case for the factory pressing of the stator bar. The master curve shifting technique was used to describe the temperature dependence for both the resin-rich mica tapes in the temperature range -50C to 50C at constant degree of curing. When the degree of curing increased the fit to one single master curve also increased. At the same time the activation energy in the frequency direction decreased and in the amplitude direction became zero. The change in dielectric response for different degrees of curing was also expressed with a master curve for both tapes and at constant temperature. Master curves were calculated for three different samples that were cured under the Factory, Soft and Hard temperature paths to the same degree of curing. It was not possible to separate these three master curves because they coincide. The conclusion from this is that it is not possible with a dielectric response measurement to say anything about the temperature path curing the sample but only about the final reached degree of curing.
174
Chapter 7
A network model with dispersive components was introduced as a way to model the dielectric response from the two resin-rich mica tapes at a fixed degree of curing and at constant temperature. Similar types of materials as in the two tapes were found in literature or measured separately. These materials were put together in two different models, a S-model and a H-model. The materials were put in series in the S-model and in a series and parallel combination allowing in-plane conduction and polarisation in the H-model. For the uncured resin-rich mica tape with woven glass as carrier material, the in-plane conduction was dominant, H-model, which was interpreted as an epoxy-resin movement through the carrier material. For the fully cured tape, this in-plane conduction became less important and the total dielectric response could be well described with a S-model. There was a strong barrier effect for the uncured resin-rich mica tape with polyester film (PET) as carrier material. Both the S- and H-model describe this barrier effect well but the Hmodel allows epoxy-resin movements around the mica flakes. The low frequency loss peak seen in tan is from the barrier effect whereas the other loss peak is related to dipolar effects in the epoxy-resin and the PET film. For the fully cured tape, both the S- and H-model work well. With this network model, it is possible in a simple way to separate between the geometrical effect and the effect from each materials dielectric response on the total dielectric response of the composite. The geometrical structure of the carrier material in the two tapes influences strongly the total dielectric response.
175
Conclusions
Chapter 8
Conclusions
8.1
Analysis of dielectric measurement methods
Three different dielectric measurement methods and the relations between them have been evaluated. The three measurement methods were in time domain polarisation/depolarisation currents and recovery voltage and in frequency domain capacitance and tan as function of frequency. These methods were applied to three different insulation systems oil/paper, resinrich mica tape and XLPE. A good way is to choose diagnostic method and domain according to the shape and magnitude of the dielectric response function in the system and according to the time/frequency range a certain interesting phenomena is expected to be found in. For the recovery voltage method, the forward and inverse problems were discussed. The forward problem to calculate the recovery voltage given f(t), and is straightforward as long as the grounding time, t2-t1, is well defined. The inverse problem to calculate f(t), and from the recovery voltage is more difficult. However, the dielectric response function can be calculated by minimising the current to an analytic expression of the response function. Different methods to Fourier transform data between the time and the frequency domains have been studied. When transforming data from the time to the frequency domain it is important to compensate the dielectric response function for the finite charging period. Otherwise, a spurious peak in frequency domain will appear for the lower frequencies. The extrapolation of data outside the measured range is critical. Especially when data has rapid time dependence, steeper slope than t-1 for short times. A simple method to use is the Hamon approximation, which works well if data approximately follows the Curie- von Schweidler model. A second method is to approximate data piecewise with cubic-splines. The Fourier transform can then be solved analytically but extending data outside the measured interval is still critical. A small error in the approximation of the dielectric response function for short times seriously affects the real part (cos-transform) of the Fourier transform. A third method is to chose base functions which have analytic Fourier transforms that can be fitted to measured data in both time and frequency domains. The advantage is that an analytic base function directly fulfils Kramers-Kronig relations. To use analytic base functions give good flexibility in handling the asymptotic behaviour of data outside the measured region. If the user has some extra knowledge about the behaviour outside the measured data range, it can be added as an extra base function.
177
Chapter 7
8.2
Study of resin-rich mica tape insulation
Two different resin-rich mica tapes have been studied. The fist tape had woven glass as carrier material and the second had polyester film (PET). For the woven glass tape, there is a drastic shift of the dielectric response, at a constant temperature, from higher towards lower frequencies as the degree of curing increases. The magnitude of tan decreases in the studied frequency window, 10 mHz to 1 MHz, rapidly in the beginning up to a degree of curing of 0.6 after which it flattens out. This is interpreted as that the final structure of the epoxyresin network is reached around 0.6 and that for higher degrees of curing more cross-links are only added to make the structure more rigid. There is also a tendency to a high frequency peak, >1 MHz, which is seen for the lower temperatures, <0C. This peak is also shifted towards lower frequencies as degree of curing increases. The dielectric response was also measured during different temperature paths for the resinrich mica tape with woven glass. These measurements show that the dielectric response change with degree of curing and with temperature. Fundamental features for all temperature paths are than an increase in temperature will lead to an increase in dielectric response and rate of cross-linking. The increased degree of curing at constant temperature will decrease the dielectric response. It was assumed that the rate of cross-linking would affect the final quality of the insulation in terms of built in tension and adhesion strength between layers and towards the conductor. To verify this three different temperature paths were chosen, Factory, Soft and Hard that would give the same final degree of curing. However, mechanical breakdown test of the adhesion between layers and towards the conductor showed only a small tendency towards a higher breakdown stress for the Soft temperature path. The assumption made about the relationship between rate of cross-linking and mechanical properties could not be verified. The dielectric response and the temperature gradient were measured during pressing of a stator bar in the factory process. The dielectric response showed the same features, especially the moving ridge in tan, as the parallel-plate samples pressed in the laboratory. Capacitance and tan measurements gave a finger print of the pressing/processing stage of the stator bar. Features like loss of horizontal pressure in the press could be identified in the capacitance measurement. The second resin-rich mica tape had a polyester film (PET) as carrier material. For the lower degrees of curing two peaks were seen in tan compared to one for the tape with woven glass. This second peak believed to originate from the barrier effect due to the solid PET film. When the degree of curing is increased, the dielectric response is shifted towards lower frequencies. An Arrhenius activated reaction model was used to model the epoxy-resin part in the two resin-rich mica tapes with woven glass and polyester film (PET) as carrier materials. The parameters in this model were different for the two tapes and were determined from differential scanning calorimetry (DSC) measurements on uncured pieces. Correlation between measured and calculated degrees of curing was good for both tapes. With a reaction model for the epoxy-resin part, it was possible to calculate how the degree of curing changed for an arbitrary temperature path. Measured capacitance and tan as function of time during a temperature path could with the help of the reaction model be re-calculated as a function of degree of curing. 178
Conclusions
Chapter 8
A minima is reached in tan at f=50 Hz during the pressing process under a Factory temperature path that can be correlated to a tan value at room temperature. Such estimation like this can be useful as a help to process the insulation in the press so that the insulation for sure will reach a certain tan value at room temperature outside the press. The final reached tan value at room temperature outside the press is often used as a quality criterion. However, caution must be taken when calculating the degree of curing when there is a temperature gradient in the sample, which was the case for the factory pressing of the stator bar. Master curves were calculated for three different samples that were cured under the Factory, Soft and Hard temperature paths to the same degree of curing. All three master curves coincided. This shows that it is not possible with a dielectric response measurement to say anything about the temperature path leading to a certain degree of curing. A network model with dispersive components was introduced as a way to model the dielectric response from the two resin-rich mica tapes at a fixed degree of curing and at constant temperature. With this network model, it was possible in a simple way to separate between the geometrical effect and the effect from each materials dielectric response on the total dielectric response of the composite. The geometrical structure of the carrier material in the two tapes influences strongly the total dielectric response.
179
Future work
Chapter 9
Future work
Throughout the work with this thesis, a few ideas have come up which were interesting but a bit out of the scope for this project. In order to simulate the factory process better than today in the test cell it must be possible to pre-program a pressure path that will run in parallel to the temperature path. It is also necessary to be able to measure and control the applied pressure more precise. With a dielectric response measurement, it is possible to say something about the current state of the insulation but not anything about how this state was reached. However, the path reaching the final state is believed to be important especially for mechanical properties. Therefore, more in depth studies concerning the change in mechanical properties when using different temperature and pressure cycles in the manufacturing process is needed. It would be interesting to continue working with the modelling of the two types of resin-rich mica tapes studied here. To study each individual material in the two tapes would give more accurate input data to the H- and S-models. Especially how the epoxy-resin changes with degree of curing and temperature is important to understand. It would then be possible with the network models to calculate the dielectric response as function of both frequency and time from a given temperature path curing the insulation. Further, it would be even more interesting to try to solve the inverse problem of finding via the network model the degree of curing measuring the dielectric response. From measurements of temperature in the press during the factory process, it was found that there is a temperature gradient in the insulation. When processing thicker insulation this temperature gradient can become a problem. With models for specific heat and heat conduction, the temperature distribution in the insulation during curing can be calculated. From the temperature distribution a degree of curing distribution can be calculated and then via the network model the total dielectric response from the whole insulation. In the real factory process, there are several other materials involved in the pressing stage like the outer semiconductive tape and the strand insulation. These materials should also be investigated and added to the network model.
181
Bibliography
Chapter 10
10
[1] [2] [3]
Bibliography
M. Abramovitz, I. A. Stegun: Handbook of Mathematical Functions, Dover, New York, ISBN 0-486-61272-4, 1972. V. Adamec, J. H. Calderwood: Electrode Polarization in Polymeric Dielectrics, IEEE Transactions on Electrical Insulation, pp. 205-214, Vol. 24, No. 2, April 1989. J. Adamov, J. Busa, A. Krivda, "Utilization of Time Constants of Polarisation Spectrum - Determined from Discharge and Return Voltage - for Estimating State of HV Rotating Machine Insulation", pp. 169-172, 7th ISH Technische Universitt Dresden, Aug 26-30, 1991. S. Amari: Evaluation of Cauchy Principal-Value Integrals using Modified Simpson Rules, Appl. Math. Lett. Vol. 7, No. 3, pp. 19-23, 1994. S. Amari, M. Gimersky, J. Bornemann: Imaginary Part of Antennas Admittance from its Real Part using Bodes Integrals, IEEE Transactions on Antennas and Propagation, pp. 220-223, Vol. 43, No. 2, February 1995. S. Amari, J. Bornemann: Efficient Numerical Computation of Singular Integrals with Applications to Electromagnetics, IEEE Transactions on Antennas and Propagation, pp. 1343-1348, Vol. 43, No. 11, November 1995. A. Anton: New Developments in Resin Rich Insulating Systems for High Voltage Rotating Machines, Proc. of Electrical Insulation Conference and Electrical Manufacturing & Coil Winding Conference, pp. 607-618, Rosemount, Illinois, USA, September 22-25, 1997. G. Arfken: Mathematical Methods for Physicists, Third Edition, International Edition, Academic Press Inc., ISBN 0-12-059810-8, 1985. N. W. Ashcroft, N. D. Mermin: Solid State Physics, Saunders College Publishing, ISBN 0-03-083993-9, 1976. A. Audoli, C. Mergiot: Use of Dielectric Tests as an Aid for Hydro Generator Maintenace, IEEE International Conference on Electrical Insulation, pp. 3-6, Toronto, Canada, 1990. A. Audoli, J. L. Drommi, Analysis of Partial Discharges Measurements and Generator Technology Evolution, Proceedings of the 3rd International Conference on Properties and Applications of Dielectric Materials, pp. 687-690, July 8-12, 1991, Tokyo Japan. C. A. Balanis: "Advanced Engineering Electromagnetics", John Wiley & Sons Inc., New York, USA, 1989. G. Blom: Sannolikhetsteori och statistikteori med tillmpningar, Fjrde upplagan, Studentlitteratur, Lund Sweden, ISBN 91-44-03594-2, 1989. A. Bognar, L. Kalocsai, G. Csepes, E. Nemeth, J. Schmidt: "Diagnostic Tests of High Voltage Oil-Paper Insulating System (In Particular Transformer Insulation) 183
[4] [5]
[6]
[7]
[8] [9] [10]
[11]
[12] [13] [14]
Chapter 10
Bibliography
using DC Dielectrometics", CIGRE, 15/33-08, Paris, 26th August 1st September, 1990. [15] A. Bognar, G. Csepes, L. Kalocsai, I. Kispal: "Spectrum of Polarization Phenomena of Long Time-Constant as a Diagnostic Method of Oil-Paper Insulating Systems", Proc. Of the 3rd International Conference on Properties and Applications of Dielectric Materials, pp. 723-726, July 8-12, Tokyo, Japan, 1991. R. Brammer, D. Rudolfsson, K. Bengtsson: Improved MICAREX for High Voltage Rotating Machines, Nordic Insulation Symposium, pp. 339-346, Bergen, Norway, June 10-12, 1996. R. Brtsch, J. Allison, T. Thaler: Factors Determining Cost and Quality of Electrical Insulation in the VPI-Process, Conference Record of the 1996 IEEE International Symposium on Electrical Insulation, pp. 259-262, Montreal, Quebec, Canada, June 16-19, 1996. D. K. Cheng: "Field and Wave Electromagnetics", Addison Wesley, Second Edition, ISBN 0-201-12819-5, 1989. R. V. Churchill, J. W. Brown: "Complex Variables and Applications", Fifth Edition, McGraw-Hill International editions, ISBN 0-07-010905-2, 1990. R. Coelho, P. Jestin: On the Return-Voltage Buildup in Insulating Materials, IEEE Transaction on Electrical Insulation, pp. 683-690, Vol. EI-22 No. 6, December 1987. P. Debye: Polar Molecules, The Chemical Catalog Company, New York, 1929. F. T. Emery: The Application of Conductive and Stress Grading Tapes to Vacuum Pressure Impregnated, High Voltage Stator Coils, IEEE Electrical Insulation Magazine, pp. 15-22, Vol. 12, No. 4, July/August 1996. Eurotherm controls: Models 2408 and 2404 PID controllers, Installation and operation handbook, Eurotherm Controls Ltd., Faraday Close, Durrington, Worthing, West Sussex BN13 3PL, UK, Issue 6 May 1997. E. M. Fort: Rotating Machine Insulation, IEEE Transactions on Electrical Insulation, pp. 137-140, Vol. 25 No. 1, February 1990. H. Frhlich: Theory of Dielectrics: Dielectric Constant and Dielectric Loss, Oxford at the Clarendon Press, 1949. U. W. Gedde: Polymer Physics, Chapman & Hall, London, UK, ISBN 0-41262640-3, First Edition, 1995. B. K. Gupta, I. M. Culbert, "Assessment of Insulation Condition in Rotating Machine Stators", Ontario Hydro, IEEE Transaction on Energy Conversion, pp. 500-508, Vol. 7, No. 3 Sept 1992. U. Gfvert, B. Nettelblad: Measurement techniques for dielectric response characterisation at low frequencies, Nordic Insulation Insulation Symposium, Lyngby, Denmark, Paper No. 7.1, 1990. U. Gfvert, E. Ildstad: "Modeling Return Voltage Measurements of multi-layer insulation systems", Proc. 4th International Conference on Properties and Applications of Dielectric Materials, pp. 123-126, July 3-8, Brisbane, 1994. U. Gfvert: "Condition assessment of Insulation Systems", Invited lecture, Nordic Insulation Symposium, pp. 1-20, Bergen, Norway, June 10-12, 1996. 184
[16]
[17]
[18] [19] [20] [21] [22]
[23]
[24] [25] [26] [27]
[28]
[29]
[30]
Bibliography [31]
Chapter 10
B. V. Hamon: An Approximate Method for Deducting Dielectric Loss Factor from Direct-Current Measurement, Proc. Inst. Elec. Eng. (IEE), Vol. 99, Monograph no. 27, pp. 151-155, 1952. J. Harwood: Teflon AF A new polymer for electronics, 8th International Electronic Manufacturing Technology Symposium, pp.503-507, IEEE, New York, USA, 1990. H. Hata, T. Maeda, N. Tagawa: Verification of Insulation Test for Traction Motors, Electrical Engineering in Japan, pp. 90-97, Vol. 129, No. 2, 1999. S. Havriliak Jr., S. J. Havriliak: Dielectric and Mechanical Relaxation in Materials: Analysis, Interpretation and application to polymers, Carl Hanser Verlag, Mnchen, Germany, ISBN 1-56990-186-4, 1997. A. Helgeson: Berkning av Dielektrisk Respons Funktion frn tervndande Spnning, Examensarbete (Diplomawork), A-EEA-9408, Kungl. Tekniska Hgskolan, Inst. fr Elkraftteknik, Avd. Elektriska anlggningar, Augusti 1994. A. Helgeson, U. Gfvert: Calculation of the Dielectric Response Function from Recovery Voltage Measurements, Proc of the 1995 Conference on Electrical Insulation and Dielectric Phenomena, pp. 97-101, Virginia Beach, Virginia, USA, 1995. A. Helgeson: Dielectric Properties of Machine Insulation studied with Dielectric Response, Licentiate thesis, Kungl. Tekniska Hgskolan, TRITA EEA-9704, ISSN 1100-1593, Stockholm, Sweden, 1997. A. Helgeson, U. Gfvert: Dielectric Response Measurements in Time and Frequency Domain on High Voltage Insulation with Different Response, International Symposium on Electrical Insulating Materials, pp. 393-398, Toyohashi, Japan, September 27-30, 1998. A. Helgeson, U. Gfvert: Dielectric Response During Curing of a Resin-Rich Insulation System for Rotating Machines, Proc. of the 1999 Conference on Electrical Insulation and Dielectric Phenomena, pp. 289-292, Austin, Texas, USA, October 17-20, 1999. A. von Hippel: Dielectric Materials and Applications, Technology press of M.I.T and J. Wiley & Sons, New York, Second printing, 1958. A. von Hippel: Dielectrics and Waves, John Wiley & Sons, New York, Second printing, 1959. S. Hvidsten, E. Ildstad, B. Holmgren, P. Werelius: Correlation Between AC Breakdown Strength and Low Frequency Dielectric Loss of Water Tree Aged XLPE Cables, IEEE Transaction on Power Delivery, pp. 40-45, Vol. 13, No. 1, January 1998. S. Hvidsten: Nonlinear Dielectric Response of Water Treed XLPE Cable Insulation, PhD Thesis, Norwegian University of Science and Technology, Dept. of Electric Power Engineering, Trondheim, Norway, ISBN 82-471-0433-4, 1999. B. Holmgren: Dielectric response, breakdown strength and water tree content of medium voltage XLPE cables, Licentiate thesis, Kungl. Tekniska Hgskolan, TRITA EEA-9705, ISSN 1100-1593, Stockholm, Sweden, 1997
[32]
[33] [34]
[35]
[36]
[37]
[38]
[39]
[40] [41] [42]
[43]
[44]
185
Chapter 10 [45]
Bibliography
E. Ildstad, U. Gfvert, P. Thrning: "Relation Between Return Voltage and Other Methods for Measurements of Dielectric Response", IEEE International Symposium on Electrical Insulation, pp. 25-28, Pittsburgh, PA USA, June 5-8, 1994. E. Ingelstam, R. Rnngren, S. Sjberg: Tefyma, Handbok fr grundlggande teknisk fysik och matematik, Tredje upplagan, Sjbergs frlag, Helsingborg, Sverige, ISBN 49-070444-0, 1986. Ingenjrshandboken: I. Allmna delen, Edited by C. A. Strberg, Second printing, Nordisk Rotogravyr, Stockholm, Sweden, 1947. Ingenjrshandboken: Allmn del/ Tekniska grundvetenskaper, Edited by S. H:son Tidestrm, Second extended edition, Nordisk Rotogravyr, Stockholm, Sweden, 1953. J. D. Jackson: Classical Electrodynamics, Third edition, John Wiley & Sons, Inc., New York, ISBN 0-471-30932-X, 1998. D. W. Johnson: A Fourier series method for numerical Kramers-Kronig analysis, J. Phys. A: Math. Gen., pp. 490-495, Vol. 8, No. 4, 1975. A. K Jonscher: "Dielectric Relaxation in Solids", Chelsea Dielectrics Press, London, UK, ISBN 0-9508711-0-9, 1983 A. K. Jonscher, L. Levesque: Volume Low-Frequency Dispersion in a SemiInsulating System, IEEE Transactions on Electrical Insulation, pp. 209-213, Vol. 23, No. 2, April 1988. A. K. Jonscher, M. A. Chaudhry, T. C. Goel: Contact and Surface Impedances on Humid Mica and Glass, IEEE Transactions on Electrical Insulation, pp. 397-407, Vol. 23, No. 3, June 1988. A. K Jonscher: "Universal Relaxation Law", Chelsea Dielectrics Press, London, UK, ISBN 0-9508711-2-5, 1996. Jones Stroud Insulations, Product information Novofilm 3/1S, Ref. LFNSFS/4/94, Longridge, Preston, PR3 3BS, UK. T. Josefsson: Formel och tabellsamling i matematik, 7th edition, Studentlitteratur, Lund, Sweden, 1982. B. Jnsson, R. Kullberg: Fysik, Gas,Vtska, Fast fas, Studentlitteratur, Lund, Sverige, ISBN 91-44-25551-9, 1986. K. Jnsson, D. Rudolfsson: Diagnostic Test of Insulation (DTI), A Test Package to Determine the Condition of Generator Stator Winding Insulation, International Conference on Large High Voltage Electric Systems, 11-11, Proc. of 31st session, CIGRE, Paris, France, 1986. A. Kaindl, R. Rckelein, J. Grindling, M. Gehrig: Simulation of the curing process upon epoxy-resin casting of electrical apparatuses a flexible way to do this, International Symposium on Electrical Insulating Materials, pp. 277-280, September 27-30, Toyohashi, Japan, 1998. Keithley: Low Level Measurements, Keithley Instruments, Inc., Revised Third Edition, Printed June 1984, Cleveland, Ohio, USA. Keithley: Model 6517 Electrometer: Users manual, Keithley Instruments, Inc., Test instrument group, Doc. No.: 6517-900-01 Rev. C, Third printing March 1995, Cleveland, Ohio, USA. 186
[46]
[47] [48] [49] [50] [51] [52]
[53]
[54] [55] [56] [57] [58]
[59]
[60] [61]
Bibliography [62]
Chapter 10
Y. J. Kim, J. K. Nelson: Assessment of Deterioration in Epoxy/Mica Machine Insulation, IEEE Transactions on Electrical Insulation, pp. 1026-1039, Vol. 27 No. 5, October 1992. K. Kimura, Progress of Insulation Ageing and Diagnostics of High Voltage Rotating Machine Windings in Japan, IEEE Electrical Insulation Magazine, pp. 1320, No.3 Vol.9 1993. C. Kittel: Introduction to Solid State Physics, Six Edition, J. Wiley, New York, ISBN 0-471-87474-4, 1986. D. Knig, I. O. Vlase: Simulation of Curing process of Epoxy Systems aiming at Manufacturing Faultless Solid Insulation, Nordic Insulation Symposium, pp. 347354, Bergen, Norway, June 10-12, 1996. L. D. Landau, E. M. Lifshitz, L. P. Pitaevskii: Electrodynamics of Continuous Media, 2nd Edition, Course of Theoretical Physics Volume 8, ButterworthHeinenann, Oxford, UK, ISBN 0-7506-2634-8, Reprinted 1998. H. M. Li, R. A. Fouracre, B. H. Crichton: Some Considerations for Obtaining the Low Frequency Response of Dielectrics from Absorption Current Measurements, Proc. of the 1994 Conference on Electrical Insulation and Dielectric Phenomena, pp.238-243, Arlington, Texas, USA, October 23-26, 1994. H. M. Li, R. A. Fouracre, B. H. Crichton: Transient Current Measurement for the Detection of Water Tree Growth in Polymeric Power Cables, IEEE Transactions on Dielectrics and Electrical Insulation, pp. 866-874, Vol. 2, No. 5, October 1995. B. Lindberg: Kompendium i tillmpad numerisk analys, Fr fortsttningskurserna vid KTH, Numerisk Analys och Datalogi, KTH, version 7, 1991. A. Maffezzoli, A. Trivisano, M. Opalicki, J. Mijovic, J. M. Kenny: Correlation between dielectric and chemorheological properties during cure of epoxy based composites, Journal of Material Science, pp. 800-808, Vol. 29, No. 3, February 1994. G. M. Maistros, C. B. Bucknall: Modeling the Dielectric Behaviour of Epoxy Resin Blends During Curing, Polymer Engineering and Science, pp. 1517-1528, Vol. 34, No. 20, October 1994. Maple V Release 5, Windows version 5.00, Nov. 27, 1997, Waterloo Maple Inc., 57 Erb Street W, Waterloo, Ontario N2L 6C2, Canada. Matlab: Optimization Toolbox Users Guide, Andrew Grace, The Math Works Inc. Cochituate Place, 24 Prime Park Way, Natick, Mass. 01760, December 1992. J. C Maxwell: A Treatise on Electricity & Magnetism Vol. 1, Unabridged third edition, Dover Publications Inc. USA, ISBN 0-486-60636-8, 1954. J. C Maxwell: A Treatise on Electricity & Magnetism Vol. 2, Unabridged third edition, Dover Publications Inc. USA, ISBN 0-486-60637-6, 1954. W. McDermid, Insulation Systems and Monitoring for Stator Windings of Large Rotating Machines, Manitoba Hydro, IEEE Electrical Insulation Magazine, pp.7-15, No.4 Vol.9 1993. F. I. Mopsik: Precision time-domain dielectric spectrometer, Rev. Sci. Instrum., pp. 79-87, Vol. 55, No. 1, January 1984. 187
[63]
[64] [65]
[66]
[67]
[68]
[69] [70]
[71]
[72] [73] [74] [75] [76]
[77]
Chapter 10 [78]
Bibliography
F. I. Mopsik, "The Transformation of Time-Domain Relaxation Data Into the Frequency Domain", IEEE Transactions on Electrical Insulation, pp. 957-964, Vol. EI-20 No. 6, December 1985. F. I. Mopsik: Stability of a Numerical Laplace Transform for Dielectric Measurements, IEEE Transactions on Dielectric and Electrical Insulation, pp. 3-8, Vol. 1, No. 1, February 1994. E. Nemeth: Measuring voltage response: a non-destructive diagnostic test method of HV insulation, IEE Proc.-Sci. Meas. Technol., pp. 249-252, Vol. 146, No. 5, September 1999. B. Nettelblad: Dielectric Properties of Liquid-Impregnated Porous Solids, PhD Thesis, Physics Department, Chalmers Tekniska Hgskola, Gteborg, Sweden, ISBN 91-7197-274-9, 1996. P. Paloniemi: Evaluation of Insulation Capability in Electrical Power Equipment: Approaches. Methods and Problems, Invited lecture, Nordic Insulation Symposium, pp. 0.0:1-0.0:16, Lyngby, Denmark, 1990. R. H. Rehder, R. E. Draper, B. J. Moore: How Good is You Motor Insulation System?, IEEE Electrical Insulation Magazine, pp. 8-13, Vol. 12, No. 4, July/August 1996. P. H. Reynolds: Field Testing Instrumentation, IEEE Transactions on Electrical Insulation, pp. 107-110, Vol. 25 No. 1, February 1990. P. H. Reynolds, B. H. Ward, An Alternative to DC Test Equipment for Performing Off-Line Diagnostic Testing on Large Rotating Machines, Proc. of the 20th Electrical Electronics Insulation Conference, pp. 47-49, Boston, USA, October 7-10, 1991. U. Ringstrm, L-E, Selin: Vgrrelselra, Akustik, Optik, Fifth edition, Teknisk Hgskolelitteratur i Sthlm AB THS, SE-100 44 Stockholm, Sweden, 1989. von Roll Isola, Product information Samicatherm 366.28, Ref. E200.1222/AV, February 1991, The Swiss Insulating Works Ltd., CH-4226 Breitenbach, Switzerland. B. K. P. Scaife: Principles of Dielectrics, Revised edition, Claredon Press, Oxford, ISBN 0-19-8565577 1998. M. R. Spiegel Advanced mathematics for engineers and scientists, Schaums outline series, McGraw-Hill Inc, 22nd printing, USA, ISBN 0-07-060216-6, 1993. M. R. Spiegel Mathematical handbook of formulas and tables, Schaums outline series, McGraw-Hill Inc, 28nd printing, USA, ISBN 0-07-060224-7, 1992. G. C. Stone: The Statistics of Aging Models and Practical Reality, IEEE Transactions on Electrical Insulation, pp. 716-728, Vol. 28, No. 5, October 1993. Storm, Nyberg, Dahlin, Bergquist: Materiallra och verkstadsteknik, Alb. Bonniers Boktryckeri, Stockholm, Sweden, 1959. G. Strang: Introduction to applied mathematics, Wellesley-Cambridge Press, Box 812060, Wellesley MA 02181 USA, ISBN 0-9614088-0-4, 1986. S. Strm: Elektromagnetisk Fltteori (EA 123), TRITA-TET-KOMP 92-2, Institutionen fr Teoretisk Elektroteknik, KTH, Stockholm, Sweden, 1992. 188
[79]
[80]
[81]
[82]
[83]
[84] [85]
[86] [87]
[88] [89] [90] [91] [92] [93] [94]
Bibliography [95] [96] [97] [98]
Chapter 10
S. Takeishi, S. Mashimo: Dielectric relaxation measurements in the ultralow frequency region, Rev. Sci. Instrum. 53(8), pp. 1155-1159, Aug. 1982. B. Tareev: Physics of Dielectric Materials, Second printing, English translation, Mir Publisher, 1979. Tettex Instruments AG: Polarisation spectrum analysis for diagnosis of insulation systems, Booklet 29, TT 200102A-05.91 P. Thrning, P. Werelius, B. Holmgren, U. Gfvert: High Voltage Dielectric Response Analyser for Cable Diagnostics, Conference on Electrical Insulation and Dielectric Phenomena, pp. 745-750, Pocono Manor, USA, 1993. P. Thrning: Watertree Dielectric Spectroscopy, Licentiate thesis, Kungl. Tekniska Hgskolan, TRITA-EEA-9703, ISSN 1100-1593, Stockholm, Sweden, 1997.
[99]
[100] C. L. Tucker III: Computer Modeling for Polymer Processing, Hanser Publishers, Munich Vienna New York, ISBN 3-446-14704-7, 1989. [101] L. Ullemar: Funktioner av en variabel, Studentlitteratur, Lund, Sweden, ISBN 9144-09001-8, 1972. [102] K. Velten, D. Schattauer: A Computer Experiment to Determine the Impregnability of Mica Tape Based Insulation, IEEE Transactions on Dielectrics and Electrical Insulation, pp. 886-891, Vol. 5, No. 6, December 1998. [103] K. Velten, A. Zemitis, F.-J. Pfreundt, E. Henning, D. Schattauer: A Method for the Analysis and Control of Mica Tape Impregnation Processes, IEEE Transactions on Dielectrics and Electrical Insulation, pp. 363-369, Vol. 6, No. 3, June 1999. [104] P. Werelius, U. Gfvert: High voltage dielectric response analyser for cable diagnostics in field conditions, Proceedings of the Ninth International Symposium on High Voltage Engineering, Graz, Austria, Paper 5661, 28 Aug. - 1 Sep. 1995. [105] P. Werelius, PhD Thesis, Dept. of Electric Power Enginering, Kungl. Tekniska Hgskolan, Stockholm, Sweden, To appear. [106] S. Westerlund, L. Ekstam: Capacitor Theory, IEEE Transactions on Dielectrics and Electrical Insulation, pp. 826-839, Vol. 1 No. 5, October 1994. [107] S. Wu, S. Gedeon, R. A. Fouracer: The measurement and Modeling of the Dielectric Response of Molecules During Curing of Epoxy Resin, IEEE Transactions on Electrical Insulation, pp. 409-417, Vol. 23, No. 3, June 1988.
189
Appendix A:
Orientational polarisation
Appendix A: Orientational polarisation

The classical treatment of orientational polarisation, free floating dipoles, under a static electric field was first done by Debye 1912 [21], [51], [88]. Assume a set of N non-interacting dipoles each freely floating in a non-polar liquid with a dipole moment p, at temperature T and local electric field E. The total energy of a dipole is made up of thermal energy plus electrostatic energy and can be written as We = k B T pE cos(
) [J ]
(A1)
where kB is Boltzmanns constant and the angle between dipole and electric field. The resulting torque the dipole is subjected to is
( ) = We
= pE sin(
) [J ]
(A2)
The average interaction energy at a static electric field of the set of N non-interacting dipoles can be calculated using Boltzmanns distribution law
pE
pE cos( ) e
0
pEcos ( k BT
sin( )d
e
0
pEcos ( k BT
sin( )d
[J ]
(A3)
which was calculated by Debye to

pE pE pE k B T = pE pE k T k T coth = pE L B B
[J ]
(A4)
where L(x) is the Langevin function L(x ) = coth(x ) x7 1 x x 3 2x 5 + + ... x 3 45 945 4725 (A5)
Assuming now that the interaction energy is much smaller than the thermal energy, which is the same as saying that the static electric field E is small, then the average interaction energy can be written as
( << k BT pE
(pE )2
3k B T
[J ]
(A6)
From this the mean induced dipole moment is calculated as

p = pE E
E
[J ]
(A7)
and the orientational polarizability for static fields as

orient
p E
p2 3k BT
[Fm ]
2
(A8)
191
Orientational polarisation
Appendix A:
To get a feeling for how the torque induced by the electric field affects the system compared to the chaotic rotational motion due to the thermal energy assume a static electric field E=108 V/m and a dipole moment p=1 D =3.335610-30 Cm. The assumed electric field is very high and corresponds to the breakdown field in solid dielectrics and the dipole moment is normal, compare p=1.84 D for H2O. From equations (A1), (A3) and (A4) the mean value of cos can be calculated as
cos E =108 V/m pE 30 = L k T p=3.335610 Cm 0.03 B T =300 K
(A9)
This shows that the alignment in the direction of the static electric field is very small even at very high macroscopic fields. It is more correct to say that an electric field induces a tendency to alignment for the molecules.
192
Appendix B:
Analytic solution to the Kramers-Kronig relations for the CvS model
Appendix B: Analytic solution to the KramersKronig relations for the Curie-von Schweidler model
Here a suggestion for the analytic solution to the Kramers-Kronig relations for the Curie-von Schweidler model is presented. This solution is not unique and follows mainly an outline for the general solution of the Kramers-Kronig relations given by Jackson [49]. The complex part of the complex electric susceptibility, (), is (
)= A
m-1
0< m <1
(B1)
Then to find the real part of the complex electric susceptibility the following integral has to be solved
)=
+a xm x (x ) 2A lim 2 lim 2 dx = a a x 2 x 0 0 +a
dx
(B2)
y R C
r -
r + x
Figure C1
Contour C used for the integration of f(z).
By using the residue theorem which says that [19]
f (z )dz = i2 Res f (z )
C k =1 z =z k
(B3)
where C is a positively oriented simple closed contour, within and on a function f(z) is analytic except for a finite number n of singular points. These singular points have each the residue Res f(zk), k=1..n. In Figure C1 the pole at z= is included and the pole at z= is excluded from the contour C. The theorem in (B3) will then give [89]
zm z2 C
(z + dz = i2 lim z
zm (z + )(z
= i )
m-1
e i m
(B4)
Then set up the eight integrals along the contour C.
193
Appendix B:
zm z2 C
dz =
0 xm ( + r ei ) ir e i d + R x m dx + + dx 2 2 ( + r e i )2 2 x2 2 x r + r 14 4 244 3 14444 4 244 3 4 244444 3 14 -r m z =x z = + r e i z =x
(R e ) iR e d + (x e ) dx + (e ( + r e )) ir e d + + (x e ) (e ( + r e )) (R e )4 1444 24444 3 1444 2444 3 1444442444443

2 i m + r i2 m i2 i m i i i 2 2 i2 2 2 i2 i 2 2 0 R 2 z = R e i z = x e i2 z=
+ r e i e i2
(x e ) dx + (e r e ) ir e d = i + (e r e ) (x4 )4 e 44 14 4 2 3 1 4444 4 244444 3

r i2 m 0 i2 i m i r i2 2 2 i2 i 2 2 2 z = x e i2 z = re i ei2
m-1
e i m
(B5)
Let now R, r0 and r0, which gives
xm x2 0
( e i2 m x m + r ei ) dx + lim ir e i d + dx + 2 2 2 i 2 r 0 x ( + r e ) 1444442444443
0 0 m I m m -1
(e i2 ( + r e i )) ir e i d = i + lim 2 i2 r 0 e i )) 2 ( r + 2 (e 1444444 4 24444444 3

II
ei
(B6)
The two last limits, I and II, can be studied using the LHospital rule [101]
lim
r 0
f (r , g(r ,
) = lim df (r , ) dr ) r0 dg(r , ) dr
(B7)
Assume now also that the limit and integration can change order. The first limit, I, can then be written as
f (r , I = lim r 0 g (r ,
0
)d )
f (r , = lim r 0 g (r ,
0
)d )
= lim
df (r , r 0 dg (r ,
d =
0
) dr d ) dr
m
= lim
r 0
m ( + r ei
m-1
e i ir e i + ( + r e i 2( + r e i ) e i
ie i
ie i i d = i 2 e 2
m 1
(B8)
For the second limit, the same approach can be used and it can be written as
II = lim e i2 m
r 0 2
f (r , g(r ,
)d )
=K =
i 2
m 1
e i2 m
(C9)
If now (B8) and (B9) are put into (B6)
(1 e ) x x
i2 m 2 0
m 2
dx (1 e i2 m )
i 2
m 1
= i
m-1
e i m
194
Appendix B:
xm x2 0
dx = i
m-1
(1 2 + 1 2 e e ) = (1 e ) 2
i2 m i m i2 m
m 1
m 1
(1 cos( P )) = sin ( P )
(B10)
P tan 2
0 < m <1
This will then finally give
) = 2A lim a
0
+a
xm x2
( (
dx = A
m 1
P tan = ( 2
0< m <1
) tan
P 2
(B11)
) = cot )
P 2
195
Appendix C:
Numerical Fourier transform by approximating data with cubic splines
Appendix C: Numerical Fourier transform by approximating data with cubic splines

Start with the Fourier transform of the dielectric response function
( f
) = f (t ) e -i
dt = f (t ) e -i t dt
0
(C1)
Since it is experimentally impossible to measure the dielectric response function for all times t > 0, the Fourier integral has to be divided into three parts
( f
) = f (t ) e -i t dt + f (t ) e -i t dt + f (t ) e -i t dt
0 1 42 4 43 4 Start approximation t1 1 42 4 43 4 Fit to cubic splines End approximation
t1
t2
(C2)
t2 1 42 4 43 4
where first and third parts represent times where there exist no measured values of f(t). These two parts have to be approximated in such a way that they contribute with as small errors as possible. With these approximations of the dielectric response function, the Fourier transform will only be valid in the following frequency region
1 < t2 < 1 t1
(C3)
The first step is to find a good approximation of the dielectric response function f(t) for times t<t1. This can be done by fitting analytic functions like a 3rd order polynomial or a Curie-von Schweidler model (t-n, 0<n<1) to the first decade of measured data and then solve that part analytically [1], [90]. The start approximation with a 3rd order polynomial can be written as
-i t 3 2 -i t f (t ) e dt = At + Bt + Ct + D e dt = 0 0 t1 t1
e = i
-i t1
3 3A 2 2B 6A C 2B 6A At1 + B + i t 1 + C + i 2 t 1 + D + i 2 i
(C4)
and the start approximation with the Curie-von Schweidler model can be written as
2 3 ( ( i t) i t) dt = ( ) f t e dt A t e dt A t 1 i t L = = + + 2! 3! 0 0 0 (C5) 2 3 m 1-n 2- n 3-n 4 -n m +1-n t1 ( ( t1 t1 i ) t1 i ) t1 m +1 (i ) = A i + + L = A ( 1) 3! 4 n 2n 2! 3 n m! m + 1 n m =0 1 n t1 -i t t1
-n
-i t
t1
-n
For the Curie-von Schweidler model it is only possible to use 0<n<1 otherwise the integral will be singular and series will converge fast if t1<1. Another way, which will be described below, is to use Richardson extrapolation [69]. The measured data will be extrapolated to t=0 and then cubic splines will be fit to the data, between t=0 and t=t2. 197
Numerical Fourier transform by approximating data with cubic splines f(t) f(a) f(a+h) f(a+2h) Measured data Estimated data
Appendix C:
t=0 Figure C1
a t1 =a+h
a+2h
Time
The approximation of the dielectric response function f(t) for times t < t1 is made with Richardson extrapolation.
The Taylor expansion of f(t), around f(a) shown in Figure C1, can be used in the following way h2 h3 f (t ) + f (t ) + ... 3! 2! 4h 2 8h 3 f (a + 2h ) = f (t ) + 2hf (t ) + f (t ) + f (t ) + ... 2! 3! f (a + h ) = f (a ) + hf (t ) + this gives the following expression for f(a) f (a ) = 2f (a + h ) f (a + 2h ) + h 2 f (t ) + 6h 3 f (t ) + ... 2f (a + h ) f (a + 2h ) 3! (C7)
(C6)
if h is small. This kind of elimination is called Richardson extrapolation and can be performed in several steps until the desired precision is obtained. When f(a) is obtained then f(a-h), f(a2h), ..., f(0) can be calculated in the same way. The next step is to calculate the cubic splines between t=0 and t=t2 [69]. It has here been chosen to fit one spline between t=0 and t=t1 and ten splines per decade from t=t1 to t=t2. The cubic splines, which are third order polynomials, are expressed as
s i (x ) = f (t i-1 ) + x (f (t i ) f (t i-1 )) + h i x (1 x )[(k i-1 d i )(1 x ) (k i d i )x ]

x= t t i-1 f (t i ) f (t i-1 ) , h i = t i t i-1 , d = i = 1,2,... t i t i-1 hi
(C8)
where ki and ki-1 are related to si(x) in the following way

d s i (x = 0 ) = h i-1k i-1 dx d s i (x = 1) = h i k i dx
(C9)
To be able to determine ki it must be assumed that the second derivative of si(x) is continuous in ti+1 and ti. This means that the following condition must be satisfied
h i+1k i-1 + 2(h i + h i+1 )k i + h i k i+1 = 3(h i+1d i + h i d i+1 ) i = 1,2,..., m 1
(C10)
where m is the total number of splines. There are still not enough conditions to solve for ki so the natural thing is to assume that the first and last spline will continue linearly out of the interval t=0 to t=t2. This gives the following two extra conditions 198
Appendix C:
Numerical Fourier transform by approximating data with cubic splines (C11)
h i+1k i-1 + 2(h i + h i+1 )k i + h i k i+1 = 3(h i+1d i + h i d i+1 ) i = 1,2,..., m 1

Now ki can be found by solving the following linear equation system
2 h 2 0 M 0 0
1 2(h1 + h 2 ) h3 M L L
L L 0 h1 0 L 2(h 2 + h 3 ) h 2 0 M M M L 0 hm L L 0
3d1 L 0 k0 k 3(h 2 d1 + h1d 2 ) L 0 1 M L 0 M = M M M M 2(h m1 + h m ) h m 1 k m 1 3(h m d m-1 + h m-1d m ) 3d m 1 2 km
All the splines are, after solving this linear system of equations, then determined. The next step is to find a function that will end the measured time domain data in such a way that as small error as possible is introduced in the frequency range studied. The easiest way to do this is to assume an ending of the form [78] f (t ) = A + B e t t > t2 (C12)
The constants A, B and can be found by solving the non-linear equation system [69]
f (t 2 ) = A + B e t 2 0.875t 2 f (0.875 t 2 ) = A + B e f (0.75 t ) = A + B e 0.75t 2 2
(C13)
When these steps have been performed the Fourier transform of the measured time domain data, f(t), can be written as
( f = s1 ( x ) e
0 1 -i t1x
) = f (t ) e -i t dt + f (t ) e -i t dt + f (t ) e -i
0 t1 t2 m 1 -i ( x ( t n - t n -1 )+ t n -1 ) n=2 0
t1
t2
dt =
t 1dx + s n (x ) e
-t -i t e dt A + B e (t n t n -1 )dx + t2
(C14)
which is an expression of the Fourier transform of f(t) which has a analytic solution. The last integral, the end approximation, can be written as
t2
(A + B e ) e
t
-i t
dt =
A i e i
(1
B +i
e (1
+i )t 2
(C15)
However there can be some numerical instabilities when integrating powers of x with sinus and cosines for small x. Study the following example
2 x sin(x )dx = 2x sin(x ) + 2 x cos(x ) = 2
9x 4 + x5 12
( )
(C16)
The integrand is (x3) for small x (sin(x)x) and because of this the integral must be (x4). But the analytic solution of the integral has a leading term of (1). This means that the first two terms in the expansion of the analytic solution to the integral cancel each other. This will cause a numerical instability for small x if the analytic form of the solution to the integral is used. This can be avoided by Taylor expansion of the sine term before integration. 199
Numerical Fourier transform by approximating data with cubic splines
Appendix C:
This method described above is a powerful tool for transforming time domain measurement data into the frequency domain and vice versa. The algorithm is both numerically stable and easy to implement in a computer.
200
Appendix D:
The time domain solution to the network model
Appendix D: The time domain solution to the network model

Here one possible way of solving the total dielectric response for a network of dispersive materials in the time domain will be presented. The solution is simplified by looking only at a H-network like that in Figure D1. However, the technique to solve this problem is the same independently of how many nodes the network has.
x1 e1 x2 e2 Z2 Z1 Z3 x3 e3 e5 Z5 U u(t) e4 Z4 i(t)
x4=0
Figure D1
A H-network with potentials xi defined in each node and voltage drops ei over each dispersive material Zi. Total applied voltage over the network is u(t).
First, all the potentials xi are defined at independent nodes and one node is defined to have zero potential. All voltage drops over each dispersive material including the driving voltage, which is applied over the network, are defined. The voltage drops can then be written in matrix and vector forms as
0 e1 (t ) u (t ) 1 e 2 (t ) 0 1 0 x 2 (t ) e = e 3 (t ) = 0 1 1 e(t ) = b(t ) - A x(t ) x t ( ) 3 e 4 (t ) u (t ) 0 1 e (t ) 0 0 1 5
(D1)
where e is the vector with all voltage drops, b the vector with all external sources which can be both voltage and current sources, matrix A represents the distribution of the nodes and x is the vector with the potentials. Knowing the voltage drop, ei(t), over the material Zi will give the current through the material as, see equation (3.21)
i i (t ) = h (e i (t )) = C 0 e i (t ) +
0
de i d t + f (t dt dt 0
)e i ( )d
(D2)
where C0 is the geometrical capacitance, is conductivity, high-frequency component of the relative permittivity and f(t) the dielectric response function of the material. From a numerical point of view, it turns out to be better to use charge rather than current. The charge movement in the material, assuming totally discharged material from the beginning, can be written as
201
The time domain solution to the network model q i (t ) = i i ( )d =

0 t
Appendix D:
= C0
0 0
e (
i
t + f (t )d + e i (t ) e ( 0 ) i { 0 =0
)e i (
(D3) )d lim f (t )e i ( )d t 0 0 14 44 2444 3 =0

t
The first integral over voltage can be approximated using the trapezoidal formula with step size t as [89]
e ( )d
i 0
W (e i (0) + 2e i ( W ) + ... + 2e i (t - W ) + e i (t )) 2
(D4)
The second integral, which is the convolution integral, can be approximated in the same way with the same step size t giving
f (t - )e ( )d
i 0
W (f (t )e i (0) + 2f (t - W )e i ( W ) + ... + 2f ( W )e i (t - W ) + f (0)e i (t )) 2
(B5)
Putting the results from equations (D4) and (D5) into equation (D3) will give the following approximation of the charge
q i (t ) C 0 W e i (0 ) + e i ( W ) + ... + e i (t W ) + 2 0 f (t )e i (0 ) + C0 W + f (t - W )e i ( W ) + ... + f ( W )e i (t - W ) + 2
(D6)
W + C0 + f (0 ) e i (t ) = r+ 2 0 = C 0 W X( e i (0 ),..., e i (t W ), f ( W ),..., f (t W )) + C 0 Y(
, f (0 )) e i (t )
The charge movements through each material can also be written in matrix and vector froms as
q1 q 2 C 1 0 W X q (t ) = q 3 q 4 C 5 0 W X q 5 C 1 0 Y +
, e1 (0 ),..., e1 (t W ), f 1 ( W ),..., f 1 (t W ) + M 5 5 5 5 5 , e (0 ),..., e (t W ), f ( W ),..., f (t W )

1
, M
1 r
, f 1 (0 )
e1 e 2 L 0 e 3 q(t ) = g (t ) + C e(t ) O M 5 5 5 5 L C 0 Y , r , f (0 ) e 4 e 5
(D7)
202
Appendix D:
The time domain solution to the network model
where q(t) is the charge vector, g(t) is the vector representing the memory effect in the material from the applied voltage before time t and C is a capacitance vector representing the instant response from the material. Kirchoffs law is then used to sum up all the currents in each node to zero, which is the same as to sum up the corresponding charges to zero. This can be written as
i 1 (t ) = i 2 (t ) + i 3 (t ) q 1 (t ) = q 2 (t ) + q 3 (t ) i 4 (t ) = i 5 (t ) i 3 (t ) q 4 (t ) = q 5 (t ) q 3 (t )
This can also be written in matrix form as
(D8)
q1 q 2 1 1 1 0 0 q 3 = 0 A T q (t ) = 0 0 0 1 1 1 q 4 q 5
(D9)
where the matrix A is the same as in (D1). Equations (D1), (D7) and (D9) can be put together into a matrix/vector system of equations written as
e(t ) = b(t ) A x(t ) -1 q(t ) = g (t ) + C e(t ) C (q(t ) g (t )) = e(t ) T A q(t ) = 0
(D10)
This system can be reduced to

-1 C (q(t ) g(t )) + A x(t ) = b(t ) T A q (t ) = 0
(D11)
and finally be put together as the following linear system of equations
C -1 Q T A
A q(t ) b(t ) + C 1 g(t ) = x t ( ) 0 0
(D12)
The solution to this system gives the charges in all materials and the potentials at all unknown nodes. This system has to be solved for each time step made with any arbitrary applied voltage u(t). The coefficient matrix will always be the same but the source vector has to be recalculated since it represents the memory effects of the system. Here a fixed step size t was chosen which can be time consuming and is not necessary. However choosing another step size will affect the expression in (D6).
203
Appendix E:
List of symbols
Appendix E: List of symbols

Activation energy in x, y-directions Admittance matrix Analytic base function, parameters Angular frequency Angle between measured IM() and IC Applied step voltage Boltzmanns constant (=1.380710-23) Capacitance Capacitance, geometrical (air) Capacitance, geometrical for surface effects in the H-model Capacitance, complex frequency dependent Charge density (volume) Cooling channels, diameter Conductivity Convection, heat transfer coefficient Current density (volume) Current, capacitive Current, resistive Current, depolarisation Current, polarisation Current, during the recovery voltage period Current, measured complex frequency dependent Current, vertical in network position x,y Current, horizontal in network position x,y Density Debye temperature Degree of curing Dielectric response function Dielectric response function, calculated from spline approximation Dielectric response function, compensated for finite charging period Dielectric response function, measured Dipole moment Dynamic viscosity Electric displacement Electric field intensity Electric field intensity, local 205 Ex,y Y ms, ns, b0s, as U0 kB C C0 C0// C( ), C( ) dcc c, coat, tot J IC IR idpol(t), id(t) ipol(t), ip(t) iR(t) IA ( ) iV x, y H i x, y d D f(t) f(t)spline f(t)comp f(t)measured p D E EL (eV) (1/) (rad/s) (rad) (V) (J/K) (F) (F) (F) (F) (C/m3) (mm) (1/m) (W/m2K) (A/m2) (A) (A) (A) (A) (A) (A) (A) (A) (kg/m3) (K) (1/s) (1/s) (1/s) (1/s) (Cm) (kg/ms) (C/m2) (V/m) (V/m)
List of symbols Electric field intensity, macroscopic applied Electric polarisation Electric susceptibility Electric susceptibility, complex frequency dependent Electric susceptibility, calculated from Hamon approximation Electric susceptibility, calculated from analytic base functions Electric susceptibility, calculated from Spline approximation Electrostatic energy Emissivity, 01 Exothermal energy, total Exothermal energy, total remaining Force Frequency Grounding period, t2-t1 Heat flow Impedance, vertical in network position x,y Impedance, horizontal in network position x,y Impedance, total of the network Magnetic field intensity Magnetic flux density Matrix describing network node configuration Mean fluid velocity Permittivity (=8.8541910-12) Permeability (=1.2566410-6) Polarisability, electronic Polarisability, atomic Polarisability, orientational + atomic + electronic Potential Potential in network node x,y Recovery electric field Recovery voltage Relation between resistive and capacitive current Relative permittivity Relative permittivity, static Relative permittivity, effective (mixture) 206
Appendix E: EA (V/m) P (C/m2) e ( ) , () Hamon Analytic , Analytic Spline , Spline We (J) e (J) Qtot QT (J) Ftot, F1, F2 f tg (N) (Hz) (s)
dP, P ,P, P (W)
ZV x, y ZH x, y Ztot
H B A0 v 0 0 optical infra-red tot ux,y ER(t) UR(t) tan r s eff
() () () (A/m) (T) (m/s) (As/Vm) (Vs/Am) (Fm2) (Fm2) (Fm2) (V) (V) (V/m) (V) -
Appendix E: Relative permittivity, high-frequency component Relative permittivity, complex frequency dependent Relative permittivity, measured imaginary part Shift function from temperature T1 to T2 in x, y- directions Shift of master curve in x-direction Shift of master curve in y-direction Specific heat Surface, (i=1..4) Stefan Boltzmanns constant (=5.670310-8) Temperature Temperature, top heating plate Temperature, bottom heating plate Thermal conductivity Thickness main insulation Time Time, end charging period Time, start recovery voltage period Time, start of measurement in time domain Time, end of measurement in time domain Voltage, applied complex frequency dependent Voltage, peak Voltage drop, vertical in network position x,y Voltage drop, horizontal in network position x,y Volume ratio of material i
List of symbols ( ) , () Measured Ax,y(T1,T2) Xshift Yshift Cp dS, si, stot S-B T TT(t) TB(t) d t t1, tc t2 tstart tend
A( U Upeak eV x, y H e x, y wi
(J/kgK) (m2) (W/m2K4) (K, C) (C) (C) (W/mK) (mm) (s) (s) (s) (s) (s)
(V) (V) (V) (V) -
207
Appendix F:
Pictures of the test cell
Appendix F: Pictures of the test cell

A few pictures are shown here of the test cell used in the chemistry lab for pressing the resinrich mica tape samples under a constant pressure and a pre-programmed temperature path.
Figure F1
Picture showing the measurement setup in the chemistry lab used for pressing resin-rich mica tape samples during which the frequency domain response was measured. Everything was controlled from a PC via GPIB- and serial port communication.
Figure F2
Picture showing the test cell. Four springs were used to generate constant pressure. The outside of the test cell was electrically isolated to avoid ground loops. Cold tap water was used for cooling. 209
Pictures of the test cell
Appendix F:
Figure F3
A four-wire connection was used to connect the sample for dielectric response measurements. The whole connection was electrically screened from external noise. The protective/insulating film on both sides of the sample in the test cell was cut in such a way that it also covered the connections.
Figure F4
The lower heating plate, showing the measurement connection and the five cooling channels. Each heating plate was individually made from a solid piece of copper with one heating element, 670 W, and one thermocouple giving temperature values to the PID-regulator.
210

Analysis of Dielectric Response

Uploaded by

Copyright:

Available Formats

Analysis of Dielectric Response

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analysis of Dielectric Response

Uploaded by

Copyright:

Available Formats

Analysis of Dielectric Response Measurement Methods and Dielectric Properties of Resin-Rich Insulation During Processing

TRITA EEA-0002 ISSN 1100-1593

Anders Helgeson Stockholm April 28, 2000

Review of electrical insulation systems in high-voltage rotating machines ____________________________________ 5

Electric properties of dielectric materials _______________ 11

Dielectric measurement techniques ____________________ 33

Test cell for processing resin-rich insulation _____________ 77

Measurements on resin-rich insulation _________________ 93

Modelling a resin-rich insulation system _______________ 137

Conclusions _______________________________________ 177

Future work_______________________________________ 181

Review of electrical insulation systems in high-voltage rotating machines

Summary of the thesis

Review of electrical insulation systems in high-voltage rotating machines

Review of electrical insulation systems in high-voltage rotating machines

Review of electrical insulation systems in high-voltage rotating machines

Different groundwall insulation systems

Review of electrical insulation systems in high-voltage rotating machines

Manufacturing process of stator coils

Review of electrical insulation systems in high-voltage rotating machines

Review of electrical insulation systems in high-voltage rotating machines

Degrading mechanisms for insulating materials

Today existing diagnostic methods for high-voltage rotating machines

Off-line diagnostic methods

Review of electrical insulation systems in high-voltage rotating machines

On-line diagnostic methods

Review of electrical insulation systems in high-voltage rotating machines

Electric properties of dielectric materials

Electric properties of dielectric materials

Electric properties of dielectric materials

Electric properties of dielectric materials 1 = lim Y 0 Y

From this the electric displacement is defined as

Electric properties of dielectric materials

Electric polarisation mechanisms

Electric properties of dielectric materials

Electric properties of dielectric materials

R12 R21 W pEAcos

Electric properties of dielectric materials

Dielectric response in time domain

(Faradays law ) (Ampres law )

Vacuum displacement current

Electric properties of dielectric materials

0 14 4 4 2444 3 Slow polarisation

Dielectric response in frequency domain

Vacuum displacement current

Electric properties of dielectric materials

Electric properties of dielectric materials

t (E1f(t t 1 ) + E 2 f(t t 2 ) + E 3f(t t 3 ) )

Electric properties of dielectric materials

Dielectric response function, f(t), of the material

Electric properties of dielectric materials

log(Dielectric response function)

Curie - von Schweidler

"General response" Slope -m Slope -n

0 < m <1 0<n<2

Electric properties of dielectric materials

0< m <1 0< n<2

Electric properties of dielectric materials

Electric properties of dielectric materials

+a (x) 2 dx = lim (x) d(lnx) a x 0