Cse 24

CSE N°24
February 2022
©2022 - CIGRE CSE N°24 February 2022 1

Table of Contents
Editorial by Rob Stephen
Mitigation measures for increasing penetration of solar Photovoltaic (PV) systems in

Distribution networks
Virtual Synchronous Generator Versus Synchronous Condensers: An

Electromagnetic Transient Simulation-based Comparison
Low-cost Ultrafast Modular HVDC Circuit Breaker
Advanced Condition Monitoring of New and Upgraded Turbine Generators for Highly
Flexible Grid Demands Using Special Fibre Optic Sensors
Line Resonance Analysis Implemented in a Cloud-based Monitoring System for

Energized Power Cables – A Case Study
Comparison of electrical clearance between Japan and other countries
Eco-design as a systemic innovation method for offshore substations
Case study of dc-MMC interconnecting two HVDC lines with different grid topologies
Information Synergy: Integrating data across disciplines and functions
Distinctive Downsizing of Cone-Type Insulating Spacer for 245 kV Class GIS by

Functional Insulating Materials
Synthetic Data – A Solution to Train Diagnostic Systems for High-Voltage Equipment

without Fault-Condition Measurements
Limits of electric eld for composite insulators: state-of-the-art and recent

investigations of overhead line insulators purchased by power utilities
Connecting two synchronous areas with an HVDC link: The NordLink interconnector
Fast Frequency Response from Transmission-Connected Solar Farms: Australian

Experience

EDITO
Editorial by Rob Stephen

CIGRE Science and Engineering (CSE) - The technical magazine of
choice
Rob Stephen
CIGRE Science & Engineering Editor-in-Chief
Welcome to this, the 24th edition of CSE, a Scopus  registered (ISSN 2426-1335)
magazine which provides the following bene ts:
1. Free publication – no nancial obligation to authors;

2. Free download – maximum exposure for authors’ work;
3. Rigorous review – reviewers are selected from the most knowledgeable experts
globally by the Study Committee chairpersons;
4. Rapid publication – the time from submission to publication can be from 6-9
months depending on the revisions required.
Despite the travel restrictions due to COVID, CIGRE managed to hold a symposium in
Ljubljana as well as a regional conference in South Africa. We therefore are very
pleased to be able to publish the best paper from the South African conference as
well as best papers from each of the Study Committees represented at the
Symposium. These papers, already peer reviewed, are published in CIGRE Science
and Engineering to acknowledge the work of the authors as well as to make their
work accessible to all stakeholders.
In addition, we are pleased to publish ve papers submitted by authors that have

undergone a rigorous review and revision process by international experts.
The topics are all very relevant to the industry at present. These include renewable
penetration, HVDC breakers, condition monitoring, a case study of connection of
HVDC lines with different topologies, information synergy, and an international
comparison of electrical clearances for overhead lines.
Authored papers (not included in the “best of” selection) include synthetic data
training for diagnosis, electric eld limits for composite insulators, a case study of an
HVDC link between two synchronous systems, frequency response from solar farms,
and analysis of virtual versus synchronous generators.
Authors are from utilities, manufacturers and academia. This provides both practical
and cutting edge research results for those studying the papers.
As can be seen from the above the papers are all extremely relevant and highly
technical covering information relating to most issues relating to power delivery
globally at present.
For those at the forefront of their area of expertise who wish to publish in CIGRE
Science and Engineering it is noteworthy that the review process is extremely
thorough. Reviewers are selected by the relevant Study Committee chairperson from

the global pool of experts within CIGRE. The selection is based on their speci c
expertise in the topic covered in the paper. There is no “mass request” sent to a pool
of reviewers who may or may not be experts in the eld. All the papers published in
this edition have been thoroughly reviewed and revised often more than once. This
ensures the paper is of the highest technical integrity and quality.
If you would like to publish in this prestigious, Scopus registered, magazine please
email Rob Stephen .
All information on how to go about publishing can be found on this page.
There is no charge to publish, the magazine is free download providing the

maximum exposure to authors. Publication can occur between 6 to 9 months
depending on the revisions required.

CSE 024
Mitigation measures for increasing

penetration of solar Photovoltaic (PV)
systems in Distribution networks
AUTHORS
A. PANDARUM - Eskom Holdings SOC Ltd South Africa
Summary
South Africa is currently experiencing an increasing growth in installation of
embedded solar PV installations on the distribution grid. This is driven by a
reduction in installation and component costs coupled with increasing tariffs
resulting in an attractive nancial business case for customers who wish to reduce
their electricity bill. However, this poses a wide range of technical challenges for
Network Service Providers (NSPs). The conventional design of the distribution grid,
whereby unidirectional power ows from the generation source to the loads no
longer exists. Moreover, the lack of visibility of these installations makes it more
challenging for the development of planning and operational strategies. As it is the
responsibility of the NSP to maintain power quality, safety and network
reliability/stability for their customers, it is essential that the consequences and
potential cost implications brought about by increasing penetration of such
installations on the distribution networks is understood. This paper discusses some
of the signi cant impacts resulting from increasing penetration of solar PV on
distribution grids, mitigation measures for these impacts as well as possible cost
implications for the utility/NSP. Furthermore, recommendations are provided for
planning of future networks with increasing PV penetrations in South Africa.
KEYWORDS
PV penetration - Voltage rise - Network asset congestion - Mitigation measures
1. Introduction

From past research reports, it is evident that there is an increasing penetration of
small scale embedded generation (SSEG) being installed on the electrical grid of
South Africa. This is mainly due to decreasing installation costs as well as increasing
tariffs resulting in an attractive nancial business case for customers who wish to
reduce their electricity bills. However, as it may seem appetizing to the customer,
Network Service Providers (NSP) are not being made aware of the installations nor
are their grids ready or designed for these systems. Furthermore, as per the national
Grid Codes, it is the responsibility of the NSP to maintain power quality, safety and
network reliability and stability for their customers. It is therefore essential that
NSP’s understand the consequences and potential cost implications for increasing
penetration of SSEG’s on the distribution networks.
The technical challenges that were previously identi ed and can be associated with
high and undesirable SSEG penetration levels are:
Power ow uctuations,
Increased technical losses,
Overloading of equipment installed on the grid such as Medium Voltage (MV)/Low
Voltage (LV) transformers and cables,
Grid protection malfunction and
Voltage variation, unbalance and overvoltage.
The primary challenges of increasing penetration is generally related to the impact

on voltage i.e. overvoltage and voltage variations. These are also considered to be
the limiting factors for increasing penetration on MV/LV grids as most research
suggests.
[1] details the results for potential technical impacts of increasing penetration of PV

in distribution networks in South Africa. The work completed in this report
highlighted the differences in impact of installing increasing penetrations of PV on a
constrained and unconstrained network. The main ndings of the report was as
follows:
The most signi cant impact stems from installing PV only at the end of the
networks for both case studies,
Overvoltage, voltage rise and rapid voltage changes/voltage variations are
experienced when installing PV. These impacts are increased as the PV
penetration increases and can have detrimental effects to the equipment that is
currently installed on the network,
Technical losses are decreased when installing PV on the constrained network for
all penetrations below 75% of peak load. However, for the unconstrained network,
the losses already start to increase by 40% penetration,
Equipment loading starts to worsen after 30% penetration for the constrained
network and worsens for any PV penetration in the unconstrained network and
The acceptable hosting capacity based on voltage and thermal loading impacts
was ~15% and ~30% of peak load for the constrained and unconstrained
network, respectively. This occurs in the case where the systems are
only installed at the end of the network. Furthermore, the voltage impacts on the
constrained feeder seems to be worse than for the unconstrained feeder.
Following the results from the case studies, this research aims provide insight into
the network management methods that can be adopted to overcome/alleviate
voltage and asset congestion impacts experienced when increasing PV penetration
in distribution networks.

2. Technical solutions and mitigation measures
This section provides details about the solutions that can be implemented to
mitigate technical impacts of increasing penetration of PV on Distribution networks
and to further increase hosting capacity on a network. Many measures have been
mentioned in research however; there is a lack of comprehensive analysis of
advantages and disadvantages of them [2]. These methods are tabulated in Table 1,
where some advantages and disadvantages are mentioned, furthermore, the
placement or responsible party is also highlighted [2, 3, 4, 5]. The advantages and
disadvantages of each measure is classi ed in terms of mitigating or managing
voltage violations or asset congestion issues caused by increasing PV on Distribution
networks. These methods have been assessed on various types of networks
theoretically using simulation tools.
The mitigation measures with associated locations are illustrated in Figure 1 below.
Figure 1 - Mitigation measures with associated locations
Details about some mitigation measures from Table 1 and its implementation is
discussed below. It must be noted that the technical solutions provided in the table
are, to a large extent, able to manage voltage violations; however, they are not always
able to effectively manage asset congestion issues. As such, [5] has propose a
combination of solutions, which will also be detailed in Section 2.10 below.

Table 1 - Summary of technical solutions and mitigation measures for increasing PV on Distribution
networks [2, 3, 4, 5, 6]
2.1. Smart inverters

Conventional grid-tied inverters allowed for the installation of SSEG’s in the South
African grid are generally controlled to feed the maximum power available from the
resource (solar, wind, etc.) and disconnect from the grid during abnormal conditions
or power outages. However, as more of these systems are being connected to
distribution grids around the world, international utilities are calling for these
inverters to support grid in an active manner [7]. This is apparent in the latest IEEE
Standard 1547 for Interconnection and Interoperability of Distributed Energy
Resources with Associated Electric Power Systems Interfaces (2018); distribution
grid codes such as Hawaiian Electric Rule 14H and California Rule 21, and inverter
standards such as AS/NZS 4777.2:2015 require the inverter based SSEG’s to provide
control functions. These controls can be autonomously and/or via smart grid
communications. The de nition of these types of smart inverters can such that they
are capable of offering the following ve system functions [8]: i) integrated sensing
and measurement functions, ii) integrated system control functions (e.g. grid-forming

and regulation capabilities), iii) integrated system protection functions, iv)
distributed system stabilization functions (e.g. virtual inertia and damping
provisions) and v) integrated cybersecurity functions.
The mitigation measure recommended here makes use of function i) mentioned

above to utilise tailored internal inverter control settings to amend the output power
(either active or reactive) of the PV system. The manner in which this is implemented
internationally is dependent on the inverter and utility codes for embedded
generation. The active power settings, referred to as Volt-Watt, entails the reduction
of active power as a function of the grid voltage metered by the inverter. Similarly,
the reactive power settings, referred to as Volt-Var, entails the compensation of
reactive power also as a function of the metered grid voltage. A visual example of the
implementation of these settings is illustrated in Figure 2 below.
Figure 2 - Example of tailored Volt-Watt and Volt-Var inverter settings [5]
One of the major advantages to this solution is that it does not require any nancial
input from the utility. Furthermore, studies have proven that this solution is highly
effective at managing voltage violations, regardless of the type of network.
2.2. Installation of Behind-the-meter Battery Energy

Storage System (BESS)
This solution requires the installation of behind-the-meter BESS alongside the PV
system which set to the operation mode that maximises self-consumption. This
means when PV generation exceeds customers demand, the BESS charges using the
excess PV generation and when PV generation falls below the customers demand,
the BESS discharges to meet the demand. However, studies [2, 8] show that with this
operation mode, the typical residential sized BESS can reach State-of-Charge (SOC)
during the day due to the systems not being adequately discharged. For that reason,
the effectiveness of managing the voltage and asset congestion issues on the network
is reduced, especially in comparison to the other mitigation measures.
2.3. Adjustment of Off-load Tap Changers in existing

transformers
This is a simple approach to creating voltage headroom in an MV network. It involves
manually decreasing the tap positions of all existing MV/LV transformers to the
lowest possible point so that additional headroom is created for more export of PV
generation until voltage violations are reached. [9] showed that this solution can

create ~5% more headroom when all taps on a network is reduced as seen in Figure
3. The main disadvantage of this solution is that it does not assist with asset
congestion in any way. Furthermore, voltage drops due to excessive loading
conditions at peak periods can become problematic. Therefore, it is recommended
that detailed network studies be completed prior to implementation, especially for
networks in South Africa, which are known to be severely loaded.
Figure 3 - Change in customer voltage that is created when adjusting all the off-load tap positions of
MV/LV transformers on a network [5]
2.4. Installation of Low Voltage Regulator (LVR)

This solution considers the installation of a LVR or Step Voltage Regulator (SVR) at an
optimal location along the LV network. The study [2] modelled the Paci c Volt LVR-
30 device (30kVA on single phase and 90kVA on three phase) which can regulate the
LV voltage by 13%, at most. The optimal location suggested by [2] was just after the
MV/LV transformer in a location that results in a step-down in voltages due to the
load of the customers on the network. This works better for urban LV networks rather
than rural networks. Unfortunately, asset congestion issues will not be solved using
this solution. Furthermore, LVR’s are not in full commercial operation yet.
2.5. Network Augmentation (MV & LV) – Transformer

upgrades and conductor reinforcement
This solution involves the uprating of existing MV/LV transformers to ones with the
largest possible rating as well as the replacement of existing high impedance MV and
LV conductors to that with larger ampacity. This can be implemented for
the conductor/s along the main and most problematic path as a more cost effective
option. This concept is illustrated in Figure 4.

Figure 4 - Network augmentation for LV feeder (substations is the same as MV/LV transformer point)
The LV and MV conductors that could be considered for this option is tabulated in
Table 2 and Table 3, respectively.
Table 2 (left) - LV conductors to consider for augmentation of LV network [2, 4] / Table 3 (right) - MV
conductors to consider for augmentation of MV network [4]
When uprating the MV/LV transformers the upgrade of transformers with two
additional tap positions of 2.5% tap size can be considered [2].
2.6. Adjustment of Zone Substation OLTC

This solution involves reducing the target voltage of the HV/MV transformer at the
zone substation to create additional headroom for voltage rise during high export of
PV power. The results are similar to the option described in Section 2.3 and can be
used in combination with it.
2.7. Installation of MV/LV On Load Tap Changer (OLTC)

transformers
This solution considers the replacement of existing MV/LV off-load tap changer
transformers to OLTC transformers. This changes the voltage management from a
passive exercise to more of an active one that occurs closer to the load. OLTCs

typically exist on the primary side of the MV/LV transformer and are controlled by a
device called the Automatic Voltage Control (AVC) relay. This AVC relay uses voltage
measurements taken from the secondary side (Vs in gure below) of the
transformer to determine the tap change requirements and inherently the voltage
regulation. There is a delay in sending the signal for the tap change to avoid
excessive tap changes stemming from possible short period voltage deviations. This
typical voltage control logic is illustrated in Figure 5.
Figure 5 - Voltage control logic of OLTCs in transformers [8]
Study [8] has indicated that the typical operation of the OLTC as mentioned above
may not be the most suitable as the target voltage is xed and this may result in the
voltage violations at the far end of the LV network. This is especially true for uneven
loads and generation from new technologies such as PV which may cause excessive
wear and tear on the tap changers. It was therefore recommended that an adaptive
control mechanism be considered when implementing this solution. This includes
which leveraging smart meter load data to actively calculate the target voltage.
OLTCs are not currently available for transformers below a certain rating [2].
However, [5] indicated that it is possible to retro t an existing transformer by adding
the active voltage control OLTC capability to it. The ones that were considered in the
studies reviewed had ±8-10% total voltage regulation capability [2, 5, 4, 10]. This
option provides mitigation for voltage issues caused by increasing PV, however, asset
congestion issues still remain a concern.
2.8. Combination solutions [4]

No one solution is able to address both voltage issues and asset congestions issues
effectively, as seen above by the advantages and disadvantages of each mitigation
measure. Most of the solutions do mitigate voltage issues and only network
augmentation is seen to address asset congestion problems but not voltage issues.
Consequently, [4] suggested that different combinations of solutions be evaluated to
identify the most effective and economical solution at tackling both voltage and asset
congestion issues with increasing PV on the distribution network. The combination
solutions evaluated were:

1. Smart inverters + network augmentation + adjustment of off-load tap changers,
2. Adjustment of off-load tap changers + smart inverters + adjustment of zone
Substation OLTC,
3. Smart inverters with tailored Volt-Watt and Volt-Var settings + network
augmentation + adjustment of off-load tap changers,
4. Installation of MV/LV OLTC transformers + network augmentation + adjustment
of off-load tap changers + smart inverters,
5. Installation of behind-the-meter BESS + network augmentation + adjustment of
off-load tap changers + smart inverters,
6. Installation of network BESS + network augmentation + adjustment of off-load tap
changers + smart inverters,
7. Network augmentation + adjustment of off-load tap changers + smart inverters +
dynamic target voltage at Zone Substation.
All the combination solutions were modelled for i) Long Rural (LR) – 692km total
length (incl. 194km Single Wire Earth Return (SWER) conductor),ii) Short Rural (SR) –
329km total length (incl. 54km SWER conductor), iii) Long Urban (LU) – 70km total
length (no SWER) and iv) Short Urban (SU) – 30km total length (no SWER); all 22kV
with pseudo LV networks. The results showed that combination solution 3, 4 and 6
works for all networks and can allow 100% PV penetration without any violations. No
other solutions worked at increasing the PV penetration on the long rural network,
however; option 1, 2 and 5 worked for the short rural and the long urban network
until 20% and 40% PV penetration.
3. Conclusion
Various measures to mitigate/alleviate the technical impacts of increasing PV
penetration on MV and LV networks have been unpacked in this report. It is evident
that no one solution is effective at solving the major issues brought about by
increasing PV penetration on distribution networks. The mitigation measures need
to be considered in conjunction with the PV penetration limits desired. It is
important to note that adjusting regulations for inverters, such as implementing
requirements with tailored Volt-Watt and Volt-Var settings, can allow for improved
quality of supply without any costs to the NSP/utility. The results from the studies
reviewed showed that with the implementation of smart inverters of these tailored
settings, PV penetration of up to 35% can be allowed in some distribution networks.
However, other impacts of high reactive power compensation on existing equipment
must be considered and fully understood prior to this change in regulation.
There are different solutions for different types of networks. There seems to be more
effective measures that exist for urban networks rather than rural networks. Study
[5] has shown that the solutions for rural networks are also much more expensive for
the same desired PV penetration level.
Network reinforcement and OLTC transformers are considered to have the highest
net- bene t for higher PV penetrations (>60%). The network reinforcement option
also seems to be the most effective at resolving asset congestion issues; whereas the
OLTC transformers options is most effective at resolving voltage issues.
It must be understood that the “do-nothing” option can signi cantly decrease the life
of existing plant and inadvertently cost the utility more money in the long run. There
is no doubt that increasing PV penetration leads to increasing grid voltage, thermal
loading and technical losses. These impacts cause network equipment to “run

hotter” and effectively reduce their mean-time to failure. Further noting that NSP’s
have not been able to perform pro-active maintenance of equipment due to nancial
constraints.
South Africa does not currently have the aspiration of increasing PV penetration at
the distribution level due to concerns around asset congestion and voltage issues
experienced when doing so. However, it must be noted that these issues may already
be experienced with existing penetrations due to lack of visibility of installations as
well as poorly managed networks. In those cases, the mitigations measures from this
report can be explored as solutions. Furthermore, when South Africa does begin to
explore higher PV penetrations, the recommendations and mitigation measures
detailed in this report, along with applicable simulations, should be considered.
4. Recommendations
The following recommendations are proposed for a successful future utility with
increasing penetration of PV on the distribution grid [2, 4]:
1. It is imperative that the inverter standards, speci cally NRS 097-2-1, are
updated to ensure visibility on the LV network as well as promote smart inverters
with dynamic Volt-Watt and Volt-Var control.
2. Upgrade MV/LV transformers with additional buck taps in the off-load tap
changer during replacement activities.
3. Increase visibility on the LV networks, especially those that are starting to
experience increases in PV penetration. A start would be to install smart meters
at critical points on the LV networks to obtain data and to create LV network
drawings and power ow models for network planners to use in their studies.
This is predominantly for future planning activities as well as for the transition
into the eld of evaluating LV hosting capacity. Currently, the LV network has
been seen as a low priority for utilities but PV has changed this notion for
international Distribution Network Service Providers (DNSPs).
4. Consider and model other complementary mitigation measures such as the
combination solutions presented above. It was evident from the studies that no
one solution can assist with all the technical impacts introduced by increasing PV
penetration on the network.
5. Explore other operational modes for battery systems. The studies showed that
operation behind-the-meter BESS in self consumption mode did not really
mitigate the voltage issues on the network. Smarter solutions need to be
developed and implemented. This can also be informed by regulation and
incentives can be given to those who implement smarter batteries for the utility’s
bene t.
6. Consider demand control measures to alleviate the impacts of increasing PV on
the network. This could include active control of large loads in customer
premises in a smart way.
7. Consider the impact that Electric Vehicles (EVs) may have on the overall impact
in the future grid with increased penetration of PV.
Bibliography
1. A. Pandarum, “Impact of Increasing Distributed Solar PV Systems on Distribution
Networks in South Africa,” International Journal of Energy and Power
Engineering, vol. 15, no. 1, pp. 47-53, 2020.

2. S. Hashemi and J. Østergaard, “Methods and strategies for overvoltage
prevention in low voltage distribution systems with PV,” IET Renewable Power
Generation, vol. 11, no. 2, pp. 205-214, 2016.
3. ENEA Consulting, “Future grid for distributed energy,” ENEA Consulting, 2020.
4. D. Cheng, M. Mather, R. Seguin, R. Robert Broadwater and J. Hambrick,
“Photovoltaic (PV) Impact Assessment for Very High Penetration Levels,” IEEE
Journal of Photovoltaics, 2016.
5. W. Nacmanson and L. Ochao, “Advanced Planning of PV-Rich Distribution
Networks - Deliverable 5: Cost Comparison Among Potential Solutions,” The
University of Melbourne, 2020.
6. K. Kotsalos, I. Miranda, J. Dominguez-Garcia, H. Leite, N. Silva and N.
Hatziargyriou, “Exploiting OLTC and BESS Operation Coordinated with Active
Network Management in LV Networks,” MDPI, 2020.
7. S. Xu, Y. Xue and L. Chang, “Review of Power System Support Functions for
Inverter-Based Distributed Energy Resources Standards, Control Algorithms, and
Trends,” IEEE open journal of Power electronics, vol. 2, 2021.
8. Y. Xue, M. Starke, J. Dong, M. Olama, T. Kuruganti, J. Taft and M. Shankar, “On a
future for smart inverters with integrated system functions,” 9th IEEE
International Symposium on Power Electronics for Distributed Generation
Systems (PEDG), 2018 .
9. A. Procopiou, M. Liu, W. Nacmanson and L. Ochoa, “Advanced Planning of PV- Rich
Distribution Networks – Deliverable 4: Non-Traditional Solutions,” The University
of Melbourne, 2020.
10. A. Procopiou, K. Petrou and L. Ochoa, “Advanced Planning of PV-Rich Distribution
Networks Deliverable 3: Traditional Solutions,” The University of Melbourne,
2020.
11. A. Navarro-Espinosa and L. Ochoa, “Increasing the PV Hosting Capacity of LV
Networks: OLTC-Fitted Transformers vs. Reinforcements,” IEEE, 2015.
Contact Author: A. PANDARUM 

CSE 024
Virtual Synchronous Generator Versus

Synchronous Condensers: An
Electromagnetic Transient Simulation-
based Comparison
AUTHORS
S. HADAVI, S. PHU ME, B. BAHRANI - Department of Electrical and Computer

Systems Engineering, Monash University, Australia
M. FARD - Aurecon Australia Pty, Australia
A. ZADEH - Array Technologies Inc.
Summary
D ue to the observation of new stability challenges in the high penetration of
renewable energy resources in weak grids, there is a need for assets to stabilize
power systems. Synchronous Condensers (SynCons) are a well-known technology as a
system strength remediation. However, SynCons capital cost is a drawback for their
operation. Virtual Synchronous Generators (VSGs) are a new technology that can
mimic synchronous machines behaviour. Therefore, there is a need to investigate
and compare SynCons and VSGs performance in weak grid integration of renewable
energy resources. This paper investigates VSG's performance in weak grid
integration of a solar farm in a single machine in nite bus (SMIB) system. The VSG
performance is compared to a SMIB case with the utilization of a SynCon.
PSCAD/EMTDC is selected for time-domain analysis of the SMIB case, and all system
equipment is modelled in detail. Different Australian grid code tests are applied to
the SMIB system to evaluate the VSG and SynCon's impact. The simulation results
show that the VSG performs better than SynCon in the weak grid system and
improves stability.
KEYWORDS
Weak grid - Virtual Synchronous Generator - Synchronous Condenser - Network Stability - Australian
Grid Code - Grid Following - Grid Forming

1. Introduction
In the recent decade, increasing global warming and technological development in
harvesting renewable energies were suf cient to motivate almost every developed
and developing country to set a renewable energy target within 10-20 years. As a
result, increasing renewable energy installed capacity in power system networks has
become a global trend. Although the de nition of renewable energy sources can be
applied to a broad range of technologies (e.g., Hydro, Geothermal, and Biomass),
generating electricity from wind and solar can easily account for the most
predominant technologies installed worldwide, which are mainly connected via
power electronic converters. Thus, renewable energy sources can address global
warming concerns and provide environmentally friendly electricity for areas away
from power transmission infrastructure. However, they also introduce new
challenges to power system networks. Increasing the number of grid-connected
renewable energy is often referred to as renewable energy penetration, theoretically
a positive factor. However, it can become a technical challenge when the ratio of
installed traditional power plants (e.g., synchronous generators) to renewable energy
sources (such as solar farms) steadily declines.
Historically, most existing power system networks consist of synchronous generators

(such as hydropower plants and steam power plants). In power systems, network
strength and system stability are de ned based on the ability of the system to
maintain the voltage in the system [1].
Nowadays, there are countless nancial and environmental incentives to increase

renewable energy penetration in power system networks and replace conventional
synchronous generators with inverter-based technologies. As this trend gains further
attention, the mechanical rotary mass ratio to inverter technologies drops into a
smaller value. Hence, the power system network is continuously experiencing a
reduction in system inertia. A power system with low inertia is prone to higher
frequency volatility, which can become quickly unstable, mainly after contingency
events. Low inertia results in faster voltage, frequency, and angular instability in any
power system, making it hard to avoid system blackouts and interruption in the
power supply [1].
In addition to inertia, synchronous generators can provide fault current contribution

of nearly 300-400% of their nominal rating; however, grid-connected inverters used
in solar farms have relatively much lower fault contribution capability. A
combination of lower fault contribution and reduced inertia causes power system
networks to experience a low short-circuit ratio (SCR) at the point of connection of
grid-connected generators. The lower SCR shows a higher voltage sensitivity to
changes in active and reactive powers import and export [2, 3]. It is worth
mentioning that not all renewable energy sources are relaying in inverter-based
technologies. However, with signi cant land accessibility and immense sun
irradiation in a country like Australia, generating electricity from solar farms can be
the most dominant means. As solar farms implement inverter-based technology
without rotating mass, a power system network like Australia can signi cantly suffer
from a de cit of inertia in the network while its solar farm penetration rises.
From generators prospect, establishing and maintaining a stable connection to a

network with a low SCR and low inertia imposes several operational issues that limit
any generator's range of stable operation. Inverter-based generators, such as solar
farms, which are normally grid following type, require rigorous tuning of their control

loops to provide stable operation whenever connected to a low strength network due
to their inherent fast response characteristic. If the stable operation cannot be
achieved by tuning the control loops, output curtailment should be adopted to lower
plant output. Furthermore, during contingency events that can cause a signi cant
voltage drop at the inverter terminals, these solar farms may not meet grid support
requirements due to existing voltage sensitivity within the grid.
To avoid curtailing maximum power extraction from solar farms in low strength
networks, one or a combination of fast-acting solutions such as Static VAR
Compensators (SVCs), Static Synchronous Compensators (STATCOMs), Synchronous
Condensers (SynCon), and grid-forming inverters (GFMI) with synthetic inertia
capability (virtual synchronous generator) should be considered as an additive to
those plants. Generally, the Virtual Synchronous Generator (VSG) can provide
additional features that can generate revenue if the ancillary market is available
apart from increasing network stability; in contrast, the former solutions can only
provide reactive power support.
This paper analyzes and compares two leading solutions, synchronous condenser
and virtual synchronous generator (VSG). Each solution's impact is investigated on
an actual Australian network scenario when a solar farm is connected to a low system
strength network. To compare a SynCon performance and a VSG performance in a
weak grid, different test scenarios are applied on a SMIB case with the SynCon and
the VSG. The test scenarios are selected based on the Australian grid code. The tests
include the large-signal tests, such as fault ride-through capability and small-signal
tests, such as voltage reference step change. The SMIB case is considered as a very
weak grid, which the solar farm cannot operate stably without any system
remediation techniques. Finally, the simulation is done in PSCAD/EMTDC software to
evaluate the VSG and SynCon performance.
The rest of this paper is organized as follows. In Section 2, an explanation of current

challenges and solutions for the high penetration of renewable resources is
presented. Section 3 describes two common assets for system strength remediation
and summarizes their advantages and drawbacks. In Section 4, the performance of a
VSG in a weak grid is investigated and compared to a SynCon performance under
different tests. Finally, conclusions derived through the results of this study are
presented in Section 5.
2. Weak grid stability issues and grid code

requirements
Recently, the high penetration of Inverter Based Resources (IBRs) and the retirement
of old fossil fuel-based generators have introduced new power systems challenges.
The majority of IBRs are located in remote areas which are connected via long
transmission lines. The long transmission lines have high impedance, decreasing the
short circuit capacity and SCR at the connection point. A power system with an SCR of
less than three is considered a weak grid. Weak grid integration of IBRs can cause
unstable operation of RERs in a system [4, 5].
Furthermore, IBRs, typically operate in a grid-following mode. In the grid-following

mode, the IBRs follow the angle and voltage from the point of connection and inject
active and reactive power according to their setpoints. In the grid-following mode,
the IBRs are using PLLs to operate stably. However, in weak grids, PLLs may cause
instability in the IBRs operation. Increasing the number of IBRs and their interaction

with grid elements introduce new challenges in power systems. Sub-Synchronous
Resonance (SSR) linked with weak grid-connected IBRs has attracted signi cant
attention in recent years. In 2009, the rst SSR related to IBRs was observed in
ERCOT [6], Texas. The interaction of a wind farm with series capacitors causes the
SSR in the system. SSR is caused by Sub-Synchronous Oscillations (SSOs) in the
system. In North China, SSOs have been observed because of type 4 wind farms and
the weak grid [7]. Also, South Australia experienced a blackout in 2016, which
cascaded wind farms' failure due to multiple contingencies that caused the blackout.
Eight wind farms protection systems tripped them after six voltage dips. Thus, the
South Australian system faced a generation reduction of 456 megawatts (MW) in less
than seven seconds [8].
To avoid the mentioned power systems challenges, power system operators in

different countries introduce standards and grid codes. All IBRs that want to connect
to the grid must follow these grid codes. Australia is one of the pioneers in the IBRs
connection, and the Australian power grid has several weak grid points to which IBRs
are connected. Therefore, the Australian grid code includes more comprehensive
and strict rules among its counterparts [9, 10]. The National Electricity Rules (NER),
the Australian grid code, has a complete chapter for connecting a new generation to
the grid, such as IBRs [11].
Moreover, there are different reactive current injection requirements during voltage
dip or reactive current absorption for high voltage ride-through and speci c voltage
settling time for each IBR, challenging IBRs to comply with those rules. Also, IBRs
interaction with other installed IBRs in weak grids can cause stability issues.
Therefore, remediation assets to increase system strength such as SynCon and GFMI
are required. In Section 3, the SynCons and VSGs are described brie y as
remediation assets for IBRs operation in weak grids.
3. System strength remediation assets
3.1. Synchronous Condenser

A SynCon is a synchronous machine without a prime mover, considered a reborn
technology to remediate system strength challenges. Since increasing the
penetration of renewable energy decreases the inertia in power systems, SynCons
can be one solution to provide inertia. Also, SynCons can regulate voltage by
injecting and absorbing reactive power [4, 12]. In weak grids, voltage is sensitive to
contingencies. Thus, voltage regulation is an advantage of SynCons in the operation
of weak grids.
Additionally, the protection of power systems is a challenging topic, and it is not

straightforward with the high penetration of renewable energy. As the number of
IBRs increases, the conventional protection over-current scheme has become
inadequate because traditional generators, including synchronous machines, have
different fault characteristics and responses compared to IBRs. IBRs fault currents
are primarily determined by their controllers. Generally, IBRs have a limited fault
current in the range of 1.2 to 1.5pu of their nominal current, and the traditional
generators can provide around six times their nominal current. Therefore, the
conventional overcurrent protection schemes and relay are unreliable and
challenging to coordinate with IBRs [13]. Consequently, they can help IBRs to comply
with strict rules and ride through the contingencies.

SynCon has noticeable drawbacks, such as high investment costs and long lead-time.
Furthermore, SynCons installation, maintenance, and operation cost make SynCons a
costly solution for system strength remediation [14]. Also, if extra inertia via ywheel
is necessary, the costs will be increased signi cantly. In addition, massive
infrastructure and a huge rotor installation of SynCons take time. It can take at least
one year to install one SynCon unit in a system [4]. Additionally, SynCons do not have
control over inertia response, and it's a natural response of SynCons.
3.2. Virtual Synchronous Generator

Generally, a grid-forming inverter (GFMI) is a family of inverters that control the
voltage and its frequency at the inverter terminal. As re ected in grid-forming terms,
these inverters are capable of black start operation, forming and maintaining the
grid through voltage and frequency regulation at their terminal. Unlike grid-
following inverters, which track and lock their output voltage with the grid voltage,
GFMI can operate in a standalone mode or self-synchronize its output voltage with
an upstream grid by a power synchronization mechanism, such as the swing
equation of synchronous generators. Hence, grid forming inverters require voltage
and frequency setpoints, while grid-following inverters require active and reactive
power setpoints.
GFMIs in their standard forms are designed to respond based on droop functions.
Thus, standard GFMIs are unable to provide inertia. However, these GFMIs can be
equipped with synchronous generators (SG's) swing equation to provide synthetic
inertia, commonly referred to as virtual synchronous generators. Typically, GFMI can
be categorized as droop-based GFMIs, VSGs, and virtual-oscillation-control-based
GFMIs [15]. However, due to the popularity and diversity of VSG applications, this
work focuses only on VSG and its impact on system strength.
VSG is a semiconductor-based power converter, which mimics the operation of a

classical synchronous generator. It is designed to provide synchronous-machine-like
features, such as grid voltage regulation, inertia contribution, and standalone
operation [16]. In other words, despite lacking mechanical rotating mass, VSG
exhibits the same characteristic and dynamic response of synchronous generator by
means of capturing swing equation characteristics in its control system.
These features are not available in conventional grid-following inverters. Instead,

VSGs' primary control is governed by the swing equation of SGs, shown in Figure 1 to
achieve the dynamic response of a synchronous generator [17]. P0 and P are the
active power setpoint and the measured active power at a VSG PoC. Dp and J are the
damping coef cient and the inertia of a VSG, respectively. ω0 , ω, and θ are the
frequency setpoint, the internal frequency, and the phase angle is used for the Park
transformation of a VSG. Finally, s denotes the Laplace variable.

Figure 1 - The swing equation that governs the frequency control of the VSG
VSGs act as voltage-controlled sources in the power network and can increase inertia
to the network [16]. Thanks to their voltage-controlled control technology, VSGs can
be connected to an inverter-based resource (IBR) plant in weak grids to support
terminal voltage, similarly to SynCons. The installation of VSGs in a renewable
energy plant can be easily conducted by retro tting several existing GFLIs with the
VSG control, as that only requires rmware updates. Alternatively, larger VSGs with
equipped battery energy storage systems (BESSs) can be installed in weak parts of a
power system to support multiple IBRs in these areas [15]. The reactive power
amount used for regulating the IBR plant's terminal voltage can be provided by
VSGs' reactive power and voltage controls. Thanks to their voltage-controlled
operation, VSGs can offer terminal voltage support to an IBR plant in weak grids by
injecting/absorbing reactive power to/from the grid. Hence, excessive reactive power
extraction from the IBR plant for voltage control in low strength networks can be
avoided. Thus, the IBR plant has more headroom for active power control.
Moreover, VSGs have been theoretically proved to have a more robust response than
the conventional grid-following STATCOMs in weak grids [18, 19]. In addition, it has
been found that installations of VSG can enhance the grid strength viewed by nearby
IBR plants [18]. This feature makes VSG become a candidate for improving weak grid
stability challenges.
3.2.1. Advantages and Disadvantages of VSGs
This section presents VSGs' strengths and weaknesses in supporting a solar farm
connected to a weak network. The advantages of VSGs can be summarized as follows.
1. The imbalance between demands and supplies in the power system might lead to
frequency deviations. Active power injection/absorption, also known as primary
reserve, is required from the plants to support the system frequency in these
scenarios. VSG's installation in IBR plants extends the plant's active power
absorbing or injecting capability due to the BESS attached to the VSGs' DC bus,
which means the VSGs enable the plant to provide a more signi cant
contribution to the frequency support [18]. However, this function is not
available in a SynCon due to the lack of a prime mover, while it has already been
up and running with the VSG control in several industrial projects [15]. Further
researches and developments are required to enable the operations of SynCon
with BESSs for the frequency support purpose.

2. Flexible damping factor and inertia settings, compared to SynCon, are essential
for VSGs. Two crucial advantages of SynCons are inertia capability and post-
contingency damping. In a SynCon, the inertia is provided by the rotating mass,
while the damping factor is determined by the damper windings, which are
physically integrated with the SynCons during its manufacturing process.
Consequently, to change the post-contingency damping of the SynCons, a
disconnection from the system is required. Moreover, much effort is needed to
replace the rotating mass of the SynCons. Therefore, these two properties are
almost non-adjustable in SynCons. On the other hand, as the synchronous
machine model is implemented in the VSGs controlling program, most of the
VSGs properties, including the damping factor, can be easily manipulated by
modifying the corresponding parameters of the inverter's rmware [19]. Also, the
inertia of a VSG can be adjusted by modifying the total capacity of BESS
connected to the DC bus of the VSG [16]. In contrast, the inertia of SynCon, due to
its rotating mass, is a xed property.
A power system stabilizer (PSS) can be used to improve the damping of SynCons.
However, it is mainly used for suppressing the oscillations via the excitation
system. On the other hand, the inbuilt damping of the VSG can directly impact
the frequency and the output voltage angle. Thus, apart from dampening post-
contingency oscillations, the inbuilt damping factor can restrain the angle
evolution during a fault and extend the critical fault clearing time [20, 21].
Therefore, more bene ts will be brought by improving the inbuilt damping factor.
3. Besides enabling frequency support service, a BESS also allows VSGs to

contribute inertia to the grid during a frequency disturbance to reduce the
resulting rate of change of frequency and improve the frequency nadir. This
feature helps the grid tackle the frequency deviation at the beginning of the
disturbance before the fast frequency response is in service. Unlike the
frequency support services provided by GFLIs, the virtual inertial response of
VSGs is instantaneous and not degraded by the frequency/rate of change of
frequency (RoCoF) measurement delays thanks to the replication of the SG's
swing equation. Furthermore, since VSGs start tackling the frequency deviation
immediately after the disturbance, less energy is injected/absorbed to support
the grid frequency. Therefore, VSGs' frequency support services are more
ef cient than those by GFLIs [22].
4. VSG is more robust in weak grids compared to grid-following solutions.

Semiconductor-based grid-following solutions, such as STATCOMs, do not operate
effectively in weak grids due to the degradation of the phase-locked loop (PLL)
[23]. On the other hand, not relying on the PLL and operating as a voltage source
allow VSGs to perform well in weak grids. The robustness of VSGs in weak girds
has been proved in the literature [24]. Besides, adding grid-forming voltage
source converters (VSCs, e.g., VSGs) to a network consisting of multiple PLL-
based converters helps improve the whole network's small-signal and robust
stability [24, 25]. This fact is revealed by means of the impedance-based analysis
and the structured singular value methods [24, 25].
For VSGs disadvantage, only low over-current capability compared to SynCons can
be considered. The over-current limitation is the thermal limits of the power
electronic switches (IGBTs) inside VSGs. As a result, the VSGs' over-current
capability is only 20%-40% of its rated current, which is much lower than the
counterpart of a SynCon [21].

4. Performance evaluation and comparison
between VSG and SynCon Near and far fault
This section conducts a case study on a solar farm (SF) connected to a weak grid. The
case study is simulated in PSCAD/EMTDC because investigating the transient and the
dynamic response of IBRs, VSGs, and SynCons must be done in a time domain-based
simulation. Also, PSCAD/EMTDC is an accepted software for EMT simulation in
Australia, and the IBRs manufacturer must model their models in PSCAD software
and comply with the Australian grid code. Several tests, including fault ride-through
(FRT), voltage disturbance, active power reference tracking, and frequency
disturbance, are applied to investigate the system's behaviours in three different
scenarios. In the rst scenario, which is the base case, the SF is connected to a Single
Machine In nite Bus (SMIB) system without any supporting device. The SMIB model
is a weak grid model with SCR below 3, which solar farms cannot operate stably.
Optimizing the SF control parameters can be a solution to enhance the stability of
the farm. However, changes in the network, e.g., faults, line trippings, might
negatively impact the effectiveness of the optimal parameters. Hence, more robust
solutions should be considered. The SG-like assets, i.e., SynCon and VSG, are possible
candidates for improving the SF's stability. Thus, in the other two scenarios, the SF in
the rst scenario is supported by either a SynCon or a VSG connected to the SF's
medium-voltage (MV) bus. As aforementioned, the VSG can be formed by retro tting
several existing GFLIs in the SF park with the VSG control. The topologies of the
three scenarios are shown in Figure 2. The parameters of the solar, the SynCon, and
the VSG are listed in Table I. The SF, the SynCon, and the VSG are modelled in detail
in PSCAD/EMTDC. The SF and SynCon models in this study are OEM developed
models that capture a full detail of an actual product.
The control diagram of the VSG can be found in Figure 3. P and Q in this gure are
the active and reactive power measured at the terminal of the VSG. ig,abc and ic,abc
are the grid-side current and the converter-side current of the LCL lter used in the
VSG, respectively. vc is the voltage across the capacitor of the LCL lter. The PWM
outputs drive the semiconductor switches inside the VSG. All control parameters of
the VSG can be found in Appendix I.
Figure 2 - Case study topology

Figure 3 - Detailed control of the VSG
Table I - Test-bed system info
The simulation results of the tests in the test-bed SMIB are presented in the
following subsections. The improvements brought by SynCons and GFMIs on
enhancing the system strength and inertia are revealed. Performance comparisons
of the remediation assets are also detailed below.
4.1. Fault-Ride-Through
Two tests are applied on the SMIB case to compare the enhancements brought by the
SynCon and the VSG in terms of the FRT capability and the fault recovery. In the rst
test, a three-phase bolted fault, which is the most severe fault scenario, occurs at the
SF's PoC and lasts for 120 ms. The responses of the system in the three scenarios are
presented in Figure 3.
From Figure 3, the post-fault responses are signi cantly improved when either the
SynCon or the VSG is connected to the PoC. Without these ancillary devices, long
oscillatory transients in the PoC voltage, i.e., VPoC, the PoC active power, i.e. PPoC, and
its reactive power, i.e., QPoC, persist for almost 1.6 s after the fault clearance. In
addition, as shown in Figure 3(a), the spikes in VPoC exceeds 1.5 pu. These poor

transients can trigger protection devices in the system, disconnecting the SF even
though the fault is successfully cleared. These undesirable transients result from the
voltage's extraordinary sensitivity to the variations in the current in weak grids [26,
27].
Moreover, these poorly damped transients in the post-fault voltage can trigger the
SF's high-voltage ride-through (HVRT) function. This causes a mode cycling
behaviour when the SF keeps entering and exiting the HVRT mode. The transitions
between the normal operating mode and the HVRT mode of the inverters in the SF
can contribute to the inadequate transient responses.
With the presence of the SynCon, VPoC, QPoC are quickly dampened and settle 0.17 s
after the fault clearance, as shown in Figure 3(a). A voltage overshoot of 25% is
recorded in this case, which shows a noticeable improvement compared to the base
case. Moreover, thanks to the high overcurrent capability of the SynCon, a relatively
high amount of reactive power is generated to support the grid during the fault, as
shown in Figure 3(d). However, the SynCon results in the low-frequency oscillatory
transient in the power responses presented in Figure 3(b) and Figure 3(e). The
oscillation lasts for about 2 s after the fault clearance with the highest overshoot of
15 MW. The frequency of observed oscillation is around 1.6 Hz.
In the third scenario, the VSG is used for supporting the SF operation. The voltage
and the power responses are well-damped by the VSG. All the responses settle within
0.17 s after the fault clearance. The overshoot and settling time of VPoC are very
similar to those in the case where the SynCon is installed. Moreover, as depicted in
Figures 3(b) and 3(e), no power oscillation is recorded when the VSG is installed.
Because the frequency in Figure 3(f) rises above one pu after the fault clearance, the
VSG absorbs an amount of active power to tackle the frequency deviation. As a result,
the power responses are overshoot-free in this case, which shows a signi cant
improvement compared to the case with the SynCon.
In general, compared to the base case, the highest voltage overshoot reduces by a
factor of two and stably settles to the pre-fault condition with the presence of either
the SynCon or the VSG. Besides, the power overshoot and settling time in the case
with the VSG is only 1/6 and 1/12 of the counterparts in the case of the SynCon,
respectively.
The SMIB case is tested with a shallower fault to validate the robustness of the VSG.
In Figure 4, a fault occurs at 5 s and is cleared at 5.43 s, with a residual voltage of 0.5
pu. Similar to the previous case, without neither the SynCon nor the VSG, poor
transients are recorded in VPoC, PPoC, and QPoC. The peak voltage oscillation reaches
1.6 pu, which can trigger protection devices, leading to a disconnection of the SF
from the grid. Although the SynCon damps the voltage transient effectively, this
device pollutes the power responses with subsynchronous oscillations, eventually
damped after 1.3 s since the fault clearance. Finally, the VSG shows its effectiveness
in enhancing both the voltage and the power responses. The voltage and power settle
within 0.16 s with minimal oscillations and overshoots. In general, reductions of
eight times and ten times are recorded in the settling time and the overshoot of the
power response when replacing the SynCon with the VSG.
Overall, the VSG is superior in damping the poor post-fault transients, which the SF
experiences in weak grids. Furthermore, compared to the SynCon, the VSG provides
similar support in the voltage recovery while not generating the subsynchronous

oscillations in the active power injected into the grid by the SynCon in transient
processes. As a result, smooth recoveries are achieved in both the voltage and power
responses using VSG.
Figure 4 - The responses of the system to a three-phase bolted fault at the PoC: (a) the PoC voltage, (b)
the active power transfer measured at the PoC, (c) the reactive power transfer measured at the PoC, (d)
the reactive power contributed by the remediation asset, and (e) the active power contributed by the
remediation, (f) the frequency at the PoC.

Figure 5 - The responses of the system to a three-phase fault at the PoC with 0.5 pu residual voltage
during the fault: (a) the PoC voltage, (b) the active power transfer measured at the PoC, (c) the reactive
power transfer measured at the PoC(d) the reactive power contributed by the remediation asset, and
(e) the active power contributed by the remediation, (f) the frequency at the PoC..
4.2. Voltage Disturbance ( POC voltage change)

In this part, two voltage disturbance tests are conducted on the SMIB case to
investigate the impacts and bene ts of using the SynCon and the VSG in weak grids.
The PoC voltage is either raised or decreased by 0.05 pu from the steady-state value
in each test.
The test results are illustrated in Figure 5 and Figure 6. In Figure 5, at 5 s, a step fall
is applied to the POC voltage. Right after the disturbance, the voltages in all three
cases deviate from the reference. The lowest nadir is recorded in the case with the SF
only. In contrast, the SynCon and the VSG in the other two cases increase their
reactive powers, i.e., Qsupport, to tackle the disturbance resulting in shallower nadirs
in these cases. Similar behaviours are observed with the rising voltage disturbance
in Figure 6, where the SF's voltage deviation is restrained by the reactive power
compensations provided by either the SynCon or the VSG.

Moreover, without the SynCon and the VSG, small oscillations persist for 0.7 s in VPoC,
PPoC, and QPoC before the system's settlement. Although the SynCon and the VSG
perform similarly in supporting the voltage, the SynCon injects low-frequency
oscillation in the active power, whose magnitude and settling time are 5 MW and 2 s,
respectively, to the grid as shown in Figure 5 and Figure 6. These oscillations are
because of SynCon exciter interaction with the IBR controller [12]. Besides, the power
responses of the SynCon are also polluted by high-frequency oscillations, whose
frequency is 50 Hz. In contrast, the VSG's responses are oscillation-free. In
conclusion, the VSG provides the SF with an optimal voltage disturbance rejection
while enhancing the power responses' transients.
Figure 6 - The responses of the system to a step change from 1.0 pu to 0.95 pu in the PoC voltage: (a)
the PoC voltage, (b) the active power transfer measured at the PoC, (c) the reactive power transfer
measured at the PoC, (d) the reactive power contributed by the remediation asset, and (e) the active
power contributed by the remediation , (f) the frequency at the PoC

Figure 7 - The responses of the system to a step change from 1.0 pu to 1.05 pu in the PoC voltage: (a)
power contributed by the remediation, (f) the frequency at the PoC
4.3. Active Power Reference change

The previous two tests show improvements brought by the VSG in voltage
contingencies. In this section, the performance of the SynCon and the VSG is
examined in an active power step change event. For example, at 5 s, a step change
from 0.85 pu to 0.65 pu is applied on the active power reference. The responses of
the three cases are shown in Figure 7.
In general, the SF itself can smoothly transit to the new power setpoint with
negligible oscillations and overshoots. However, with the SynCon being active,
severe undamped oscillations occur in all the responses after the step change,
re ecting the system's unstable operation. The unstable behaviour of the SynCon is
because of its exciter interaction and the SF [12]. The peak-peak magnitude of the
active power oscillation is almost 80 MW. Additionally, the voltage persistently
uctuates between 1.06 pu and 0.96 pu. These responses will result in a
disconnection of the system from the grid in reality. These oscillatory behaviours are
not observed in the case with VSG activated. The VSG retains the smooth transition
of the SF and further restrain the voltage overshoot right after the step change

thanks to its fast reactive power control. This test reveals the risk of instability
caused by the interaction between the SynCon and the SF in a power setpoint
variation event, while this issue is not present if the VSG is used.
Figure 8 - The responses of the system to a step change from 0.85 pu to 0.65 pu in the active power
reference: (a) the PoC voltage, (b) the active power transfer measured at the PoC, (c) the reactive
power transfer measured at the PoC, (d) the reactive power contributed by the remediation asset, and
(e) the active power contributed by the remediation , (f) the frequency at the PoC
4.4. Frequency Disturbance

In this test, a frequency disturbance is synthesized by ramping the grid's frequency
from 50 Hz to 51 Hz at a rate of 2 Hz/s. This disturbance occurs at 5 s in Figure 8. As
a result, the active power responses in all three cases decrease to tackle the
frequency rise.
Without the SynCon and the VSG, severe oscillation is observed in the voltage
response even in the steady-state after the frequency ramp, which causes
oscillations in the SF's active and reactive power responses. The presence of the
SynCon helps dampen the voltage but, similarly to previous tests, pollutes the active
power response with low-frequency oscillations. The magnitude of the power

oscillation in the case with SynCon is 15 W peak-peak initially and diminishes to zero
after 15 s. In the case with the VSG, both the voltage and the power responses
experience smooth transients and stably settle after the disturbance. Thus, these
responses are oscillation-free in both the transient and steady-state of the system.
Moreover, thanks to its battery storage system, the VSG can absorb active power
from the grid to support the grid frequency in this test, which is not achievable with
SynCon.
Figure 9 - The responses of the system to a frequency ramp from 50 Hz to 51 Hz at a rate of 2Hz/s: (a)
power contributed by the remediation, (f) the frequency at the PoC
5. Conclusions
Due to a need for performance comparison between SynCons and VSGs in weak grid
integration of renewable energies, in this paper, the solar farm in the SMIB scenario
with low SCR at the PoC is modelled in PSCAD/EMTDC. Different Australian grid code
tests, such as frequency disturbance, voltage disturbance, reference active power
step, and fault ride-through capability, are applied in the system considering three
con gurations: with SynCon, with VSG, and without any asset. The simulation results

show the signi cant impact of VSG on solving stability issues in weak grids compared
to the system without any assets. Also, the results illustrate that VSGs can dampen
oscillations in the system after disturbances and perform better than SynCons in the
frequency disturbances and active power setpoint change. Clearly, the VSG performs
better in all tests scenarios conducted in this study.
Appendix - Control parameters of the VSG
Aknowledgment
We acknowledge FIMER contribution by providing a PSCAD solar inverter model for
the purpose of this study.
References
1. P. Tielens, and D. Van Hertem, "The relevance of inertia in power systems,"
Renewable and Sustainable Energy Reviews, vol. 55, pp. 999-1009, 2016.
2. D. Kim, H. Cho, B. Park, and B. Lee, "Evaluating In uence of Inverter-based
Resources on System Strength Considering Inverter Interaction Level,"
Sustainability, vol. 12, no. 8, pp. 3469, 2020.
3. AEMO, "System Strength Impact Assessment Guideline," 2018.
4. S. Hadavi, M. Z. Mansour, and B. Bahrani, "Optimal Allocation and Sizing of
Synchronous Condensers in Weak Grids With Increased Penetration of Wind and
Solar Farms," IEEE Journal on Emerging and Selected Topics in Circuits and
Systems, vol. 11, no. 1, pp. 199-209, 2021.
5. M. Z. Mansour, S. P. Me, S. Hadavi, B. Badrazadeh, A. Karimi, and B. Bahrani,
"Nonlinear Transient Stability Analysis of Phased-Locked Loop Based Grid-
Following Voltage Source Converters Using Lyapunov's Direct Method," IEEE
Journal of Emerging and Selected Topics in Power Electronics, pp. 1-1, 2021.
6. Y. Cheng, S. Huang, J. Rose, V. A. Pappu, and J. Conto, "ERCOT subsynchronous
resonance topology and frequency scan tool development.", 2016 IEEE Power and
Energy Society General Meeting (PESGM), pp. 1-5.
7. W. Du, Q. Fu, H. Wang, and Y. Wang, "Concept of modal repulsion for examining
the subsynchronous oscillations caused by wind farms in power systems," IEEE
Transactions on Power Systems, vol. 34, no. 1, pp. 518-526, 2018.
8. R. Yan, T. K. Saha, F. Bai, and H. Gu, "The anatomy of the 2016 South Australia
blackout: A catastrophic event in a high renewable network," IEEE Transactions
on Power Systems, vol. 33, no. 5, pp. 5374-5388, 2018.
9. H. T. Mokui, M. A. Masoum, and M. Mohseni, "Review on Australian grid codes for
wind power integration in comparison with international standards." pp. 1-6.

10. AEMC, "National Electricity Rules Version 160," AEMC, 2021.
11. T. de Atholia, G. Flannigan, and S. Lai, "Renewable energy investment in
Australia," Bulletin—March Quarter, 2020.
12. S. Hadavi, D. Rathnayake, S. G. Jayasinghe, A. Mehrizi-Sani, and B. Bahrani, "A
Robust Exciter Controller Design for Synchronous Condensers in Weak Grids,"
IEEE Transactions on Power Systems, pp. 1-1, 2021.
13. D. Tzelepis, E. Tsotsopoulou, V. Nikolaidis, A. Dysko, V. Papaspiliotopoulos, Q.
Hong, and C. Booth, "Impact of synchronous condensers on transmission line
protection in scenarios with high penetration of renewable energy sources,"
2020.
14. J. Jia, G. Yang, A. H. Nielsen, E. Muljadi, P. Weinreich-Jensen, and V. Gevorgian,
"Synchronous Condenser Allocation for Improving System Short Circuit Ratio."
2018 5th International Conference on Electric Power and Energy Conversion
Systems (EPECS). IEEE pp. 1-5.
15. D. B. Rathnayake, M. Akrami, C. Phurailatpam, S. P. Me, S. Hadavi, G. Jayasinghe, S.
Zabihi, and B. Bahrani, "Grid Forming Inverter Modeling, Control, and
Applications," IEEE Access, vol. 9, pp. 114781-114807, 2021.
16. J. Driesen, and K. Visscher, "Virtual synchronous generators." 2008 IEEE power
and energy society general meeting-conversion and delivery of electrical energy
in the 21st century. pp. 1-3.
17. D. Pan, X. Wang, F. Liu, and R. Shi, "Transient stability of voltage-source
converters with grid-forming control: A design-oriented study," IEEE Journal of
Emerging and Selected Topics in Power Electronics, vol. 8, no. 2, pp. 1019-1033,
2019.
18. Y. Yang, Q. Ye, L. J. Tung, M. Greenleaf, and H. Li, "Integrated Size and Energy
Management Design of Battery Storage to Enhance Grid Integration of Large-
Scale PV Power Plants," IEEE Transactions on Industrial Electronics, vol. 65, no. 1,
pp. 394-402, 2018.
19. X. Meng, J. Liu, and Z. Liu, "A generalized droop control for grid-supporting
inverter based on comparison between traditional droop control and virtual
synchronous generator control," IEEE Transactions on Power Electronics, vol. 34,
no. 6, pp. 5416-5438, 2018.
20. Z. Shuai, C. Shen, X. Liu, Z. Li, and Z. J. Shen, "Transient Angle Stability of Virtual
Synchronous Generators Using Lyapunov's Direct Method," IEEE Transactions on
Smart Grid, vol. 10, no. 4, pp. 4648-4661, 2019.
21. T. Qoria, F. Gruson, F. Colas, X. Kestelyn, and X. Guillaud, "Current limiting
algorithms and transient stability analysis of grid-forming VSCs," Electric Power
Systems Research, vol. 189, pp. 106726, 2020.
22. Y. Li, and B. Jones, "The Use of Extreme Value Theory for Forecasting Long-Term
Substation Maximum Electricity Demand," IEEE Transactions on Power Systems,
vol. 35, no. 1, pp. 128-139, 2020.
23. J. Z. Zhou, H. Ding, S. Fan, Y. Zhang, and A. M. Gole, "Impact of Short-Circuit Ratio
and Phase-Locked-Loop Parameters on the Small-Signal Behavior of a VSC-HVDC
Converter," IEEE Transactions on Power Delivery, vol. 29, no. 5, pp. 2287-2296,
2014.
24. C. Yang, L. Huang, H. Xin, and P. Ju, "Placing Grid-Forming Converters to Enhance
Small Signal Stability of PLL-Integrated Power Systems," IEEE Transactions on
Power Systems, 2020.
25. R. Rosso, S. Engelken, and M. Liserre, "Robust stability investigation of the
interactions among grid-forming and grid-following converters," IEEE Journal of
Emerging and Selected Topics in Power Electronics, vol. 8, no. 2, pp. 991-1003,
2019.
26. M. G. Taul, X. Wang, P. Davari, and F. Blaabjerg, "Current limiting control with
enhanced dynamics of grid-forming converters during fault conditions," IEEE

Journal of Emerging and Selected Topics in Power Electronics, vol. 8, no. 2, pp.
1062-1073, 2019.
27. S.-H. Huang, J. Schmall, J. Conto, J. Adams, Y. Zhang, and C. Carter, "Voltage control
challenges on weak grids with high penetration of wind generation: ERCOT
experience." pp. 1-7.
Contact Author: S. HADAVI 

CSE 024
Low-cost Ultrafast Modular HVDC Circuit

Breaker
AUTHORS
S. NEE, T. MODEER, L. ANGQUIST, S. NORRGA - SCiBreak AB
Summary
The vast majority of HVDC installations so far are of the point-to-point type, which
connects two points in an AC grid, or two AC grids, by a DC link. As HVDC becomes
more prevalent, it will be highly bene cial to start interconnecting HVDC links on the
DC side into multiterminal systems, and eventually, to form HVDC grids. This
reduces the number of needed converter stations radically, making HVDC more cost-
effective, as highly expensive converters are avoided. Also, power losses are
reduced. However, multiterminal HVDC requires access to reliable and fast DC circuit
breakers for fault clearing. The VSC-assisted resonant current (VARC) technology,
pioneered by SCiBreak, can combine these features at acceptable cost. Furthermore,
VARC ef ciently lends itself to modularization, whereby a breaker is assembled from
a number of identical series-connected breaker modules. This brings two important
bene ts. First, scalability in terms of voltage at minimum engineering effort is
obtained. Second, fault tolerance can be achieved by adding one or more redundant
breaker modules. In case one of the modules fail to operate, the remaining modules
can still interrupt the fault current.
Given the criticality of HVDC equipment for the electrical energy system, type testing
of components is of paramount importance. KEMA Labs in Arnhem, Netherlands
have set up an infrastructure in line with the latest standardization activities for
HVDC breaker testing, that uses low-frequency short-circuit generators to provide
fault currents similar to those in HVDC.
An 80 kV modular VARC-based breaker using three submodules with 12 kA

interruption capability has been developed and manufactured. It has been tested
under various realistic circumstances at KEMA Labs. It was found that it could
interrupt a fault current of 12 kA with 1.5 ms neutralization time and 15 kA with 2
ms neutralization time. By neutralization time is meant the delay from the order to
interrupt is given until the fault current starts decreasing.

KEYWORDS
HVDC - HVDC Circuit Breaker - Multiterminal HVDC - vacuum interrupter
1. Introduction
The transition to a decarbonized electricity supply system calls for increased use of
HVDC power transmission. The best sites for renewable energy sources are often
located far from consumers. Often long cables need to be used, for instance in the
case of offshore wind power, and here the HVDC is the only choice since AC cables
cannot feasibly be made longer than approx. 100 km. Also, the variable nature of
renewables, such and wind and solar power, makes more long-distance transmission
necessary, so that the variability can be balanced out over a larger area.
As HVDC links become more prevalent it will become favorable to connect them into
multiterminal connections, that may eventually evolve into a DC grid. This can save
investment cost, since the number of costly converter stations can be reduced. In
addition, security of supply is improved since power can be routed in alternative
paths in case of a converter fault, for instance. However, in order for this to be
possible HVDC circuit breakers (CBs) need to be available as a mature technology at
feasible cost. Development efforts to this end have been made since the 1970s.
Recent endeavors include the hybrid HVDC CB[2]. It is based on using a mechanical
contact (an ultra-fast disconnector, UFD) for conducting the current in the normal
case, whereas power semiconductors are used for interruption. While this concept
has been shown to work well in lab tests, it is rather costly, as the power
semiconductors need to be dimensioned for both the full returning voltage and the
full fault current. To reduce cost, a mechanical breaking element, such as a vacuum
interrupter (VI) can be used instead. However, it needs a zero-crossing in the current
to stop conducting and start blocking voltage. Zero crossings are normally present in
any AC circuit, but to interrupt DC an arti cial zero crossing must be provided. An LC
resonant circuit connected in parallel with the VI, where the capacitor is initially
charged to high voltage, can be used to provide a current pulse that cancels out the
line current through the VI. A type of DC breaker that relies on this solution is the
active current injection HVDC CB [3]. It can be manufactured at reasonable cost, but
interruption is comparatively slow and takes 2−8 ms. Furthermore, it requires
specialized vacuum interrupters, and a fast high-voltage switch to block the
capacitor voltage in the normal case.
In [4] a new class of DC CBs was rst presented. It also uses a resonant circuit, but in
this case it is excited by a small voltage source converter (VSC). Hence it is named
VSC assisted resonant current (VARC™) breaker. It offers several bene ts. Cost can be
kept low as the power electronics only need to be rated for a fraction of the voltage
handled by the breaker. High–speed operation is possible with proven operating
times of below 2 ms. In addition, this type of breaker can be manufactured using
mostly standard, off-the-shelf, components.
In high-voltage applications modular equipment designs, where combinations of

many submodules are used to handle high voltage and power provide several
bene ts. Scalability and fault tolerance can be achieved. Notably, the modular
multilevel converter [6] has completely taken over the market for VSC HVDC
converters during the last 15 years. The VARC concept can also be modularized,

using a number of submodules each of them being able to block a fraction of the
system voltage. This paper describes the design and testing of such a modular VARC
HVDC circuit breaker.
2. HVDC grids and the need for DC circuit

breakers
The vast majority of HVDC systems installed to date are of the point-to-point, or
back-to-back types, where two points in the same AC grid, or two different AC grids,
are connected by a DC link. Interconnecting HVDC links on the DC side into
multiterminal connections is a crucial path towards reducing costs and increasing
reliability, see Figure 1. These could eventually evolve into an HVDC grid. The
bene ts of DC-side interconnection are that the number of expensive converter
stations can be radically reduced, lowering investment costs. Furthermore, with
point-to-point connections, power passing from one DC feeder to another will be
subject to around 2 % losses in the converters during conversion to AC and back to
DC. This is avoided with the direct DC interconnection, leading to further reductions
in lifecycle costs.
HVDC grids of any considerable complexity will require DC circuit breakers to clear
faults. A DC short- circuit fault propagates rapidly, meaning that it needs to be
cleared in few milliseconds to avoid having to block the converters, which would
mean disrupted transmission in the grid. In addition, in a DC circuit the peak fault
current is not limited by the network impedance, only the current rate of rise,
implying that the DC CB has to limit the fault current before it rises to a level that is
harmful for equipment in the grid. If a suf ciently fast HVDC breaker cannot be
provided, the only option is to connect series inductance, but that causes additional
cost and power losses. A further requirement on HVDC circuit breakers is that cost
must not be excessive. Since the purpose of developing HVDC grids is to save
investment cost it is obvious that the cost of the breaker cannot approach that of a
converter station, even though it is clear that a DC breaker will be considerably more
expensive than an AC breaker of similar rating.
Figure 1 - Comparison of HVDC point-to-point connections (left), and an HVDC grid with HVDC circuit
breakers (right)

3. Modular VARCS HVDC circuit breaker
The VARC concept for lends itself well to designing scalable HVDC circuit breakers by
series connecting a variable number of modules of lower voltage capability.
3.1. Current interruption principle

The main circuit of a VARC circuit breaker, and the principal voltage and current
waveforms during a current interruption are shown in Figure 2. The line current is
conducted by metallic contacts in a vacuum interrupter (VI) during normal, closed,
conditions. Thereby, the losses are kept low, and complex cooling systems are not
needed. Furthermore, the contact system has high current carrying capability when
closed. The vacuum interrupter has good current interruption capability at current
zero- crossings and can withstand both high voltage and high voltage-derivative
following current extinction. However, an arti cial current zero cross-over in the VI
current is required for current extinction. To provide it, a current injection branch,
consisting of a capacitor C, an inductor L and a small voltage source converter (VSC),
is connected in parallel with the main VI. During an interruption, contact opening by
an ultrafast actuator is rst initiated. When the contact separation is suf cient, the
VSC is controlled to produce a high-frequency square-wave voltage, UVSC, which
results in an oscillating current with increasing magnitude IRES in the injection
branch. This current soon approaches the line current IS in magnitude. Since the
resonant circuit also comprises the VI, this implies a zero crossing in the current
through the VI, upon which the VI interrupts.
Inductance exists in all power systems, which means that the line current cannot
immediately be brought down to zero. Instead, an alternative path must be provided
for the current when the interruption through the VI is executed. The new current
path shall cause a high, preferably constant, opposing voltage that exceeds the
source voltage in the system, thereby forcing line current down to zero. Metal oxide
varistors (MOV), possess that property, and are widely used for voltage limiting in
power systems. An MOV is connected across the interrupter terminals, and its
clamping voltage is in this context called the transient interruption voltage (TIV), see
[1]. It is selected to be approximately 50% higher than the system DC voltage. The
MOV also serves the purpose of limiting the voltage across the vacuum interrupter to
a level it can withstand.

Figure 2 - VARC DC CB concept: main circuit (left), indicative waveforms (right)
3.2. Breaker Submodule Implementation

The modular HVDC breaker is assembled from submodules rated at 40 kV TIV. This
voltage is determined by the clamping voltage of the surge arrester stack connected
in parallel with the VI. Since the surge arrester is connected nearly in parallel with
the module terminals, the minimum clearances within the module are determined in
accordance with the module surge arrester voltage level, rather than the system
voltage level. The module, see Figure 4, is divided into a low voltage compartment
and an open-frame high voltage part. The low voltage compartment houses the VSC,
the control system and the charging circuits. The module design is based on a
welded steel frame, to provide enough strength for stacking. During the breaker
operating time, 1.5 ms, the VI contacts needs to be separated several millimeters.
Such rapid movement is achieved using a Thomson coil actuator, as shown in Figure
3. In addition to a fast opening time, the actuator mechanism also needs to maintain
suf cient contact pressure in the closed state, to keep the contact resistance low and
the peak current withstand capability suf ciently high.
An FPGA-based control system controls the VSC, the VI, and the charging circuits in
the module, as well as to maintain communication with other modules and the
substation controller. The VSC control requires fast sampling with low propagation
delay, both to be able to accurately switch the VSC during the resonant current
excitation and to be able to quickly detect when the module successfully has
interrupted the current. The sampling period and sampling delay are therefore both
less than a microsecond. All components used in the breaker module are commercial
off-the-shelf, which lowers the cost. The modules are designed to be stackable, with a
maximum number of six modules in a stack.

Figure 4 - Breaker submodule. The footprint is
1.4 m × 1.4 m. Each added module adds 1 m to
Figure 3 - Vacuum interrupter (top) with the height of the stack
mounted Thomson coil actuator (bottom)
3.3. Complete Modular HVDC CB with Series-connected

Submodules
A suitable number of modules are connected in series to form a full circuit breaker.
These modules all work individually, and contain everything needed to contribute
their part to the total breaker TIV during an interruption operation. At least one of
the modules works as a central control, that relays signals from the substation to
each of the modules in the string. All modules are capable of acting as a central
control module, and can do so if connected with optical bers to the substation
control system. Information about module status, measurement values from
transducers connected to any one module et cetera, can be shared freely among the
modules and the substation controller. To prevent restrikes after interruption, and
ensure even stressing of the MOVs, it is important that the voltage is shared evenly
between all modules. On a time scale of milliseconds, voltage (and power) is evenly
shared between the modules through the robust clamping of the MOVs in each
module. On a very short time scale, voltage is shared between the modules using
small snubbers connected across the vacuum interrupters of each module. In this
way, the energy and voltage can be shared evenly under all operating conditions [8].
Large variations in energy dissipation between the modules can only occur as a
consequence of failure of a module to make or break. The circuit breaker can be
made tolerant to such a failure in one or more of the modules, by a slight increase of
the MOV rating. As an example, the energy dissipation per module in a 320 kV circuit
breaker is around 30% higher if one module fails to operate than it is if all modules
operate [7].
Like other types of DC circuit breakers, VARC requires auxiliary power. The standby
auxiliary power need of each breaker module is, however, very low, in the range 10-
20 W. The auxiliary power system is dimensioned primarily by requirements on the
breaker recharge time, i.e. the time for the breaker to be ready for a repeated rapid
open-close-open sequence. Auxiliary power supply of 300 W to each module results
in a recharge time of 90 seconds, after which an open-close-open cycle can be

performed in quick succession again. If a longer recharge time is acceptable, the
auxiliary power supply per module can be reduced down to just above the standby
power requirements.
A prototype HVDC CB consisting of three modules (Figure 5), fully demonstrating the
operation of a modular circuit breaker, was designed, built, and tested as part of the
PROMOTioN project. This circuit breaker was designed for operation at 80 kV, with a
TIV of 120 kV. For the targeted 80 kV system voltage, high voltage isolation
transformers operating at 50 Hz were chosen for the auxiliary power supply. The
main ratings of this circuit breaker are compiled in Table 1.
A VARC circuit breaker for 320 kV could in straight-forward manner be constructed

by adding more modules of the same type, connecting them together and mounting
corona shields to attachment points on the module frames, see Figure 6. In addition
to inherent redundancy, a modular design allows for simpli ed testing, since nearly
all functionality can be veri ed through the testing of a single module. Since only one
module design is required, less effort is also required for component procurement
than if the breaker needed be redesigned for every voltage level.
Figure 6 - Rendering of a 320 kV HVDC breaker

with residual current switch
Figure 5 - 80 kV modular HVDC breaker

Parameter Value
Rated DC voltage 80 kV
Rated DC current 2000 A
Rated maximum fault current 12 kA
Minimum operating time 1.5 ms
Maximum dissipated energy 2.5 MJ
Table 1 - Ratings of the tested modular VARC HVDC CB
4. Testing of the modular HVDC breaker
4.1. Laboratory facilities used

For the initial in-house testing of the modular HVDC breaker a simpli ed test circuit
based on the capacitor discharge was used [9]. KEMA laboratories have made great
efforts at developing an infrastructure for realistic testing of HVDC circuit breakers.
Based on analyses of realistic fault cases in HVDC grids the proper test requirements
have been identi ed [10]. Accordingly, test facilities have been set up using low-
frequency short-circuit generators [12] to mimic the short-circuit transients in HVDC
systems. These were employed for the test of the modular HVDC breaker.
4.2. Results
Short-circuit test were carried out at various current levels, as summarized by Table
2 below. Representative waveforms from a fault current interruption test (TF100) are
shown in . The tests TC100 and TF100 represent load current switching and fault
current interruption, respectively. The test TF100* at 15 kA shows the maximum
breaking capacity of the breaker. TF100, the rated maximum fault current breaking
test, was initially performed according to speci cation, with 2 ms operation time, and
then again with 1.5 ms operation time to demonstrate the capability of the breaker.
The test TC10 illustrates low-current switching and the TDT test illustrates
interruption with long subsequent suppression time and high energy dissipation. All
tests except TF100* were made at both current directions without changes to the
circuit breaker or its settings.

Table 2 - Overview of tests performed on circuit breaker at KEMA laboratories in August 2020. All tests
except TF100* were split into different current directions, so e.g. TF100 was performed with two tests
at 12 kA and two tests at –12 kA
5. Conclusion
A full-scale modular HVDC circuit breaker of the VARC type has been designed and
tested. The tests conform that the circuit breaker allows for a neutralization speed of
1.5 ms at 12 kA interrupted current, and a maximum interruptible current of 15 kA.
It is thus proven that the modularization of the VARC concept works under realistic
HVDC fault conditions. The modular design, provides scalability, so that any multiple
of 40 kV can be chosen as the TIV of the circuit breaker. Furthermore, the design
enables fault tolerance through redundancy, as well as a reduction in component
cost and testing.
In summary: fast, inexpensive, scalable, and fault-tolerant HVDC circuit breakers are
now ready for deployment.
Acknowledgments
The work described in this paper was made possible thanks to support of InnoEnergy
Scandinavia, Svenska Kraftnät and the Swedish Energy Agency.

Figure 7 - Test results from KEMA when switching 12 kA against 120 kV returning voltage test
This work was supported by the European Union’s Horizon 2020 research and
innovation program under grant No.691714.
References
1. CIGRE Technical Brochure 683 “Technical Requirements and Speci cations of
State-Of-The-Art HVDC Switching Equipment” (Joined Working Group
A3/B4.34, April 2017).
2. Häfner J., Jacobson B.: Proactive hybrid HVDC-breakers – A key innovation for
reliable HVDC grid, Cigré symposium, Bologna, Italy, 2011.
3. Tokoyoda S., Tahata K., Kamei K., Kikuchi K., Yoshida D., Oukaili S., Yamamoto R.,
Ito H.: DC circuit breakers for HVDC grid applications – HVDC and power
electronics technology and developments, CIGRE LUND 2015, paper 135
4. Ängquist L., Norrga S., Modeer T.: A new dc breaker with reduced need for
semiconductors, EPE’16 ECCE Europe, Karlsruhe, Germany, 2016.
5. L. Ängquist, S. Nee, T. Modéer, A. Baudoin, S. Norrga, N. A. Belda ”Design and test
of VSC assisted resonant current (VARC) DC circuit breaker” (15th IET Internation
al Conference on AC and DC Power Transmission (ACDC 2019), 5-7 Feb. 2019,
Coventry, UK).
6. A. Lesnicar and R. Marquardt, “An innovative modular multilevel converter
topology suitable for a wide power range,” in 2003 IEEE Bologna Power Tech
Conference Proceedings,, vol. 3, 2003.
7. M. Poikilidis, R. Smeets, N. Belda, S. Nee, S. Mebrahtu-Melake, T. Inagaki,
"Document on failure mode analysis of HVDC circuit breakers", deliverable 10.5
of the EU Horizon 2020 project PROMOTioN, 2020.
8. S. Liu et al., "Modeling, Experimental Validation, and Application of VARC HVDC
Circuit Breakers," in IEEE Transactions on Power Delivery, vol. 35, no. 3, pp.
1515-1526, June 2020.
9. S. Nee, T. Modeer, L. Ängquist, S. Norrga and A. Baudoin, "Tank test circuit for fast
DC circuit-breakers," 15th IET International Conference on AC and DC Power

Transmission (ACDC 2019), Coventry, UK, 2019.
10. N. A. Belda, C. A. Plet and R. P. P. Smeets, "Analysis of Faults in Multiterminal
HVDC Grid for De nition of Test Requirements of HVDC Circuit Breakers," in IEEE
Transactions on Power Delivery, vol. 33, no. 1, pp. 403-411, Feb. 2018.
11. N. A. Belda and R. P. P. Smeets, "Test Circuits for HVDC Circuit Breakers," in IEEE
Transactions on Power Delivery, vol. 32, no. 1, pp. 285-293, Feb. 2017.
12. N. A. Belda, C. A. Plet and R. P. P. Smeets, "Full-Power Test of HVDC Circuit-
Breakers With AC Short- Circuit Generators Operated at Low Power Frequency,"
in IEEE Transactions on Power Delivery, vol. 34, no. 5, pp. 1843-1852, Oct. 2019.
Contact Author: S.NEE 

CSE 024
Advanced Condition Monitoring of New and

Upgraded Turbine Generators for Highly
Flexible Grid Demands Using Special Fibre
Optic Sensors
AUTHORS
T. BOSSELMANN, S. STRACK - Siemens AG, Germany

J. R. WEIDNER - GenAdvice Consulting, Germany
Summary
The fast-rising amount of renewable energy with volatile electrical power
generation results in a need of grid stabilization demands. New highly exible power
stations in combination with upgrading projects of existing units request for
improved generator design solutions tailored to these new grid requirements. In
parallel the decreasing prizes of electrical energy at the liberalized market leads to
cost pressure on the power plant assets of the utilities. This forces manufacturers to
develop a new generator portfolio with higher ef ciency, excellent reliability at all
kind of operation regime, adequate maintenance intervals and much lower lifetime
operation costs.
Advanced condition monitoring of the generator main components like stator

winding and stator core enable early indication of operational wear or accelerated
aging effects due to harsh peak load operation regime. Special designed bre optic
sensers (FOS) can be used at many locations inside the generator, that are not
accessible for conventional sensors due to high electric and magnetic eld strength.
This paper describes the application of different type of FOS used for design
validation during prototype tests and applied for permanent online monitoring of the
unit at peak load operation on grid demands.
The following bre optic sensors are examined, most of them use Fibre Brag Grating
(FBG) technology:
Distributed temperature sensor arrays directly applied to the copper conductor

of stator winding bars,

Acceleration (vibration) sensors installed at front-end position of stator winding
baskets,
Strain sensors applied at selected stator bar involutes of highly stressed phase
separations,
Magnetic ux sensors for veri cation of calculated magnetic eld distribution at
slot exits and at axial overhang locations.
An example of end-winding vibration monitoring using FOS depicts the usefulness of

permanent assessment of generator overhang components to avoid transient
overstressing and early defects of winding coils or structure elements. Permanent
monitoring of vibration modes and displacement amplitudes ensure good and safe
operation condition with well predictable long-term wear.
In a second example the need of stator winding copper temperature monitoring with
bre brag arrays is pointed out, to avoid undiscovered thermal hot spots at the high
voltage winding insulation system resulting in sudden defects with high economic
impact.
This essential information gathered from sets of FOSs applied at pre-calculated

winding locations ensures a reliable long-term operation of the end-windings even
under harsh operation conditions. Service experience with bre optic online
monitoring obtained at several machines of new or improved design demonstrate
the advantageous use of FOS for condition assessment at highly stressed generators
of nowadays.
KEYWORDS
Fibre optic sensors - bre brag grating technology - online monitoring of operational stress - grid
demand - fast load fluctuation - on-site eld validation - stator end-winding vibration - operational
temperature distribution of copper conductor, high voltage stator winding
1. Introduction
The fast-rising amount of renewable energy with volatile electrical power generation
results in a reorientation of the electric energy market [1] and a need of grid
stabilization demands [2]. New highly exible power stations in combination with
upgrading of existing units with lifetime extension request for improved generator
design solutions tailored to these new grid requirements [3]. In parallel the
decreasing prizes of electrical energy at the liberalized market leads to cost pressure
on the power plant assets of the utilities. This forces manufacturers to develop a new
generator portfolio with higher ef ciency, excellent reliability at all kind of operation
regime, adequate maintenance intervals and much lower lifetime operation costs.
For grid stabilization with reactive power demands it includes optimized design for
new state-ot-the-art synchronized rotating condensers [4].
A new pressurized air-cooled turbine generator with a direct water-cooled stator

winding meets all these requirements. It represents an innovative generator series,
which combines proven technologies with new design concepts like load-dependent
pressurized air-cooling. This machine type is an excellent solution for exible high-
power applications as well as synchronous condensing applications where additional
grid stability is needed. Its key features are presented in separate papers [3], [4].

Advanced condition monitoring of generator main components like stator winding
and stator core enable early indication of operational wear or accelerated aging
effects due to peak load operation regime. Special designed bre optic sensers (FOS)
can be used at many locations inside the generator, that are not accessible for
conventional sensors with copper leads due to high electric and magnetic eld
strength [5], [6].
2. Types of ber optic sensors applied AT

generator
For design validation during prototype Factory Running Tests (FRT) a large bunch of
all kind of sensors was installed at the generator. Table 1 lists all 1230 sensors
placed at many speci c locations within all parts of the generator to get a maximum
of information during each step of prototype running tests. Most of them will stay in
the machine for online monitoring at peak load operation in the power plant later
on to get rst information about real service stress running at the grid.
Table 1 - List of different type of sensors installed during prototype running procedure in the generator
test eld of the factory; Fibre Optic Sensors (FOS) highlighted in green
As can be seen, for FRT more than 300 FOS of different kind had been installed at the
generator and remotely online monitored from the test eld control room. This
includes the following type of bre optic sensors, most of them use Fibre Brag
Grating (FBG) technology.
Temperature Sensor: -60 … +250°C

Acceleration (Vibration) Sensors: Resolution ±0,1 m/s², range ±0,5 … 400 m/s²
Strain Sensors: Sensitivity ±1 μm/m

Proximity (Displacement) Sensors: Sensitivity ±10 μm/m, dynamic range ±10
mm/m
Magnetic Flux Sensor: H- eld sensitivity 0,1 mT, dynamic range 1 Tesla
Distributed Temperature Sensor Arrays (FBG) at stator winding bars: -60 …
+200°C
To give an overview of the physical mode of operation and technically realisation of

the mainly used sensor types, a short description with an application example at a
turbine generator is presented in the next chapters.
2.1. FBG temperature sensor arrays

Fibre Optic Temperature (FOT) sensors used at stator windings of large generators
are based on FBG principle with up to 12 sensors in series on one bre only
(temperature sensor array), that is placed near the copper conductor to measure the
distributed warming-up of the winding direct at the source [7]. In Fig. 1 the
application of FOT sensors during stator bar manufacturing process is shown. It can
be seen, that the FBG sensors are placed directly at the narrow side of the pressed
copper conductor bar before starting the taping process for applying the HV main
insulation.
Figure 1 - Array of bre optic temperature sensors applied at narrow side of copper conductor Roebel
bar before starting of HV insulation taping
Application details together with temperature measuring results gained at a

complete generator stator winding are described in chapter 4, see below.
2.2. Accelerator sensor

In order to be able to measure end-winding vibration behaviour of high voltage
stator end-windings at high potential and high magnetic ux levels a novel bre
optic accelerometer has been developed. The working principle is based on intensity
modulation of light using a simple con guration of a bare multimode bre that is
placed towards a tilted mirror [8]. A de ned length of the bre sticks out at the end of
the ferrule like a cantilever bar, as shown in Fig. 2. When the sensor head is
accelerated the bre is bended due to its own inertial force. This changes the
distance of the bre end to the mirror and therefore the amount of light back
re ected into the bre. The design avoids relatively moving parts and therefore
leads to a very tough and small sensor housings. Because of a passive operating

sensor concept using a low power LED, the accelerometer can be used in an
explosion protected area without further precautions like hydrogen cooled
generators or HV motors at chemical industry and off-shore platforms.
Figure 2 - Working principle of cantilever type bre optical accelerometer based on intensity
modulation of light reflected at a mirror and vibration sensor produced with non-metallic elements
made of insulation material
Some typical examples for a design dependent installation of Fibre Optic

Acceleration (FOA) sensors at front-end stator overhang locations are presented in
Fig. 3.
Figure 3 - Examples of FOA sensor locations at different type of stator end-winding design depending
on generator power and cooling category
2.3. FBG strain sensor

The bre optic strain sensor is suitable for monitoring dynamic strain at stator
winding bars and endwinding structure elements inside large hydro or turbine
generators. The sensor – an example is shown in Fig. 4 – consists of insulating
materials only and it is completely passive. This makes it ideal for areas where
conventional electrical strain gauges are not useable. Therefore, it is preferably used
at locations, that are exposed to electrically high potential, high magnetic elds,
severe electromagnetic interferences and explosive gas atmospheres. It can be
placed on top of a high voltage stator winding bar as presented in Fig. 4. It is also
applied at highly dynamic-mechanical stressed support elements of end-windings.
Since dynamic strain measurements do not need to be compensated for slow

ambient temperature drifts, the sensor consists of a single FBG only. Light from a
broadband light source is re ected at the FBG structures. Depending on dynamic
strain variation and quasi-static temperature behaviour, the spectral signal changes
that can be measured with a spectrometer. The sensor signals are composed of a

slowly varying direct part (coming from static strain and temperature) and a fast-
alternating dynamic part (coming from dynamic strain) that can be easily
distinguished.
Figure 4 - Fibre optic dynamic strain sensor with application examples at top of a stator winding bar
The mechanical overstressing of a stator bar high voltage insulation system, that is
mainly caused by high dynamic-mechanical transients or thermo-mechanical
stresses due to overload conditions could lead to accelerated fatigue. Therefore, it is
recommended for speci c applications of hydro or turbine generators that the strain
should be measured in addition to the vibration behaviour of the winding overhang.
Permanent installation of strain sensors had been performed at two top layer
winding bars of different electrical phase in the slot exit region of an indirect cooled
generator, see Fig. 5. Four FBG strain sensors were applied at each bar at different
axial locations (blue arrows in Fig. 5 → ) to measure the dynamic mechanical tension
of the insulation in this highly stressed part of the overhang.
Figure 5 - Tiny FBG strain sensors (small black strips marked with blue arrows) xed with white glue
and adhesive tape strips on the surface of stator winding insulation at end-winding involute – two top
bars at phase separation
2.4. Magnetic flux sensor

During prototype tests of new designed and more ef cient generator components
like stator bars or advanced core end shape the measurement of local magnetic eld
distribution is essential for veri cation of used calculation programs. Magnetic ux
sensors that could be applied at locations with high electric potential and
electromagnetic interference would be the best option for these test eld runs.
Magnetic eld sensors basing on the Faraday effect had been developed. The
realized design for practical application is presented in Fig. 6. Linear polarized light
is guided through the Faraday crystal while it rotates the polarization vector. The

modi ed light is then converted into an intensity modulation by a second polarizer
which is oriented an angle of typically 45° to the input polarization. Light intensity
modulation correlates to magnetic ux variation and can easily be measured with
optical equipment.
As shown in Fig. 6 (b) and (c) the sensor body is made of ceramic, glass prisms,
Faraday crystal, optical polarizers of glass and high temperature resistant plastics,
which makes it completely dielectric.
Figure 6 - Principle of magnetic flux sensor with a Faraday crystal (a) and practically realized bre optic
sensor (c) as small as a 1cent coin
The new bre optic magnetic ux sensors were implemented at stator winding
overhang of turbine and hydro generators [9] to measure the magnetic eld
distribution at axial locations between stepped core and front-end bar connections
as shown in Fig. 7. Because these ux sensors are made of non-metallic material, the
applied sensor and connection line could easily be positioned all over the end-
winding construction without taking into consideration any obstruction of high
electric or magnetic eld distribution as demonstrated in given image sections of
Fig. 7.
Figure 7 - Application of magnetic flux sensors at stator winding exit in stepped core area of an indirect
cooled generator
3. Example 1 - Stator endwinding vibration

monitoring with ber optic acceleration
sensors
During load operation the generator stator windings will be excited by the following
three major sources of force:

1. Bearing and shaft vibration acting at the 1st grid harmonic (e.g. 50 Hz) for 2-pole
generators,
2. Magnetic forces of rotating ux eld in stator core also called “core ovalization”
forces acting at the 2nd grid harmonic (e.g. 100 Hz) and
3. Electromagnetic forces generated by stator current in the winding with grid
frequency. The stator current generates forces with double grid frequency (e.g. 50
Hz) resulting in a vibration frequency of the 2nd grid harmonic (e.g. 100 Hz) and
higher harmonics.
Fibre optic accelerometers can easily measure the vibrational behaviour and
dynamic-mechanical stress of stator end-windings operating at high potential well
above 10 kV and even higher. Different type of sensors are commercially available on
the market and several application papers are presented so far [10], [11]. With this
important information it is expected to get better knowledge on stress dependent
changes in the end winding structure and to ensures a reliable long-term operation
even in harsh peakload condition. The following example gives an impression of
online end-winding vibration monitoring
Figure 8 - Monitoring of stator end-winding behavior during different generator load conditions at
volatile grids (a) End-winding vibration level at start-up and shut-down process of a gas turbine
generator in peak load cycle (b) Displacement of four selected stator bar front-end locations (top to
bottom bar connections) in correlation to generator power condition
Permanent online monitoring of end-winding vibration behaviour as depicted in Fig.

8(a) ensures avoiding overstressing of overhang structure during extreme peak-load
demand at volatile grids due to high content of renewable energy feed-in. The
in uence of peak-load cycle on end-winding vibration level as given in Fig. 8(a) can
be pointed out as follows:
Different rotor speed dependent low level vibration resonance peaks at rotor
speed of ~ 700 rpm during typical gas turbine leaning cycles before starting the
ramp up of a gas turbine with the generator running in motor mode,
Some normal end winding resonance peaks during transient ramp up of the rotor
to full speed before synchronization to grid,
Low vibration level of end-windings at synchronous load operation of the
generator as expected by design requirements,
During fast ramp down only one load rejection resonance peak because of
transient winding current at the moment of disconnection from grid,
No thermal equilibrium of stator end-winding reached and therefore no thermo-
mechanical in uences on winding basket resonance behavior measurable.

The second example of online vibration monitoring at eld condition is presented in
Fig 8(b). It demonstrates the distinctive in uence of active and reactive power
variation on the end-winding displacement amplitudes (0-pk) at the stator bar front-
end connections, see trend diagram of a few days as shown in Fig 8(b). As there is not
a strong and direct correlation between power and bar displacement the delaying
in uence of temperature and thermo-mechanical stresses has to be taken into
consideration.
This example clearly illustrates the sophisticated dynamic-mechanical behaviour of

the overhang structure of a high voltage stator winding. Fibre optic vibration sensors
together with other non-metallic sensors are indispensable measuring tools for
design validation tests at stator end-windings to produce a reliable and robust
overhang basket. For more details about end-winding vibration measurement with
FBG sensors and for interpretation of test results the following literature could give
further and deeper information [8], [12] - [14].
4. Example 2 - Stator winding temperature

monitoring
To get information about the thermal behaviour of generator stator windings and
possible design warm spots at stator bars in the overhang or slot section FBG sensor
arrays could be implemented for direct temperature measurement of copper
conductors that operate at high voltage [15]. At an indirect aircooled generator many
FBG temperature sensors have been installed at selected locations along the slot
and involute part of a stator winding bar is shown by the green marker points in
Figure 9.
Figure 9 - Bottom half of an indirect cooled generator stator with marked positions of FBG sensors
measuring axial distribution of copper conductor temperature
4.1. Stator end-winding area

During the prototype Factory Running Tests (FRT) of a new generator the copper
temperature distribution of a stator end-winding area has been monitored.
Therefore, ve FBG sensors were applied at selected top and bottom bars each, to
measure the axial distribution of the copper conductor temperature between the
lead end and the core end at the winding overhang. In Figure 10 the axial
temperature distribution of three top bars and two bottom bars is displayed.
Figure 10 - Copper temperature distribution of selected top and bottom bars at stator end-winding area
between lead end and core end during test eld run – indirect air-cooled generator
The following valuable information can be collected from the given temperature
readings:
The highest copper temperature is measured at the top-to-bottom bar brazing

connection at the lead ends because of eddy currents warming up at copper
pieces. Nevertheless temperature is far below thermal class temperature of the
insulation system.
The lowest temperature is – as expected from design calculation – in the radially
well cooled middlepart of the end-winding area.
The temperature increase towards core end accrues due to higher magnetic
elds close to stepped stator core end.
The average temperature of bottom bars is lower than the one of top bars
because of cold air ow direction.
The measured copper temperature distribution at stator bar envolutes enables
an excellent crosscheck with obtained results of sophisticated generator
calculation programs.
4.2. Stator winding slot part

The temperature distribution within the stator winding slot part, measured during
the test eld run of an indirect air-cooled generator is given in Figure 11. The
sensitive temperature readings of FBG arrays at the different cooling zones (radial
cold and warm air ow) give excellent data for validating the used thermal
calculation program [16]. Due to the symmetric cooling zone design of the stator core
with axial blower at drive and non-drive end, the temperature distribution is only
measured at one side up to the center of the stator core, see Fig. 11.
Figure 11 - Axially measured copper conductor temperature distribution at top and bottom layer bars
compared with conventional slot RTD reading during test eld run
The following conclusions can be drawn from the temperature distribution at

winding copper conductors presented in Fig. 11:
The copper conductor temperature at top layer bars is signi cantly higher than
at bottom layer bars because of higher eddy current losses at top bars generated
by greater magnetic ux density at upper part of the core slots.
The in uence of radial cooling zones between warm and cold air sections is only
signi cant at bottom layer bars. At top layer bars the cooling air ow is too much
in uenced by mixed air gap ow.
The bre optic slot temperature sensors (brown dots in Fig. 11) located near
conventional RTD sensor position reproduce the difference between warm and
cold cooling air zones in an excellent way. But because these temperature
sensors are placed on top of insulated copper bars the measured temperature is
about 7 K lower than the warm spot temperature measured with FBG arrays
directly at copper conductor surface without high voltage insulation in between.
The temperature reading of conventional slot RTD sensor installed according to
IEC Speci cation is about 30 K lower than the measured copper conductor hot
spot temperature of the stator winding. Nevertheless, the thermal class
temperature of the insulation system of 155°C is far away to be touched. Also, the
utilization temperature of the stator winding insulation of 130°C is not reached.

Summarized results:
The FBG temperature arrays installed directly at the copper conductor surface
produce an excellent thermal image of the local temperature distribution inside the
stator winding during generator operation. After some years of in eld operation at
different type of generators no sensor defects due to broken bre or even sensor
related temperature drift has been observed on any installed FBG sensor. During
a planned maintenance stand still the sensor readings were checked. After waiting
some time for cooling down, the reading from the FBG sensor were compared with
the conventional slot RTD and the cooling air thermometer. All values of comparable
locations within the cooled down generator stator do agree within a margin of ±1°C,
which is accurate enough for monitoring purposes.
5. Conclusions
A severe peak load operation regime of rotating machines due to volatile grid
demands results in accelerated wear of components and enhanced aging of the
stator winding insulation with increasing risk of unexpected sudden break-down.
Therefore, the exact knowledge of stator winding behaviour as one of the most
stressed parts of the generator is essential for long-term prediction of a new
designed machine. Extensive prototype Factory Running Tests (FRT) need to be
performed [17].
It has been shown in the paper, that the use of different kind of bre optic sensors
can essentially help to achieve a realistic picture of stator winding stressing. The
advantage of bre optic sensors (FOS) is that they can be placed at all critical winding
locations because they do not have any metallic and magnetic parts and therefore do
not in uence the high electrical and magnetic elds at generator stator.
A short description of different type of FOS including a sample application has been
given for the following Fibre Bragg Grating (FBG) sensors:
Distributed temperature sensor arrays directly applied to the copper conductor

of stator winding bars,
Acceleration (vibration) sensors installed at front-end position of stator winding
baskets,
Strain sensors applied at selected stator bar involutes of phase separation,
Magnetic ux sensor for veri cation of calculated magnetic eld distribution at
slot exits and axial overhang locations.
A more detailed example of end-winding vibration monitoring using FOS

demonstrates the usefulness of permanent assessment of generator overhang
components to avoid transient overstressing and early defects of winding coils or
structure elements. Field operation experience shows, that permanent monitoring of
vibration modes and displacement amplitudes ensure a good and safe operation
condition and predictable long-term wear.
In a second example the need of stator winding copper temperature monitoring with
bre bragg arrays is demonstrated, to avoid undiscovered thermal hot spots in the
high voltage winding insulation system resulting in sudden defects with high
economic impact. This essential information gathered from sets of FOSs applied at
calculated winding locations ensures a reliable long-term operation of the end-
windings even in harsh operation condition.

The service experience with bre optic online monitoring obtained at several
machines of new design demonstrate the advantageous use of FOS for condition
assessment at highly stressed generators of nowadays.
References
1. Hans-Werner Sinn: Buffering volatility: A study on the limits of Germany’s energy
revolution, European Economic Review, Volume 99, October 2017, pp. 130 – 150
2. ENTSO-E Completing the map – Power system needs in 2030 and 2040 – Status
November 2020 · Version for public consultation, ENTSO-E, Rue de Spa 8, 1000
Brussels, Belgium
3. Hendrik Steins: Design validation, operational data assessment and new
maintenance concepts - using the example of the new air pressurized SGen-
2000P technology, 14. Essener Tagung - Generatoren in konventionellen
Kraftwerken, Windparks und Wasserkraftwerken, HdT, February 2020, Essen,
Germany
4. M. Baca, K. Walli, A. Joswig: Proper design and con guration of synchronous grid
stabilizers to cover the wide variety of requirements, IEEE PES General Meeting,
August 2018, Portland, Oregon, USA
5. T. Bosselmann, M. Willsch: Optical Sensors for Harsh Environments – Product
Portfolio – Sensors for Power Generators & Gas Turbines, Corporate Technology -
Siemens AG, 2017
6. Thomas Bosselmann: Innovative applications of bre-optic sensors in energy and
transportation, 17th Optical Fibre Sensor Conference OFS, Bruges, 23. –
27.05.2005
7. J. R. Weidner: Direct Measurement of Copper Conductor Temperature at
Generator Windings with Fibre Bragg Grating Sensors, SC A1 Rotating Machines,
Contribution PS1/Q1.6, Cigre 2012, Paris, France
8. L. Hübner, T. Bosselmann, A. Jungiewicz, G. B. Gabrielsen, E. Feilkas, K. Dybvik, H.
Daum, M. Villnow: Monitoring of winding overhang vibrations on large
synchronous motors, Petroleum and Chemical Industry Conference Europe,
Amsterdam, 6.6.2014
9. M. Villnow, M. Willsch, T. Bosselmann, B. Schmauss, T.Hildinger, T. Hess:
Distributed magnetic eld measurement on stator winding heads of a 300 MVA
water turbine motor-generator using ber-optical magnetic eld sensors, Proc.
IEEE Sensors, Nueremberg, 2010
10. Tetreault, A.: On-line End Winding Vibration Measurement Using Fiber Optic
Sensor Technology, EPRI Major Component Reliability (MCR) European Workshop,
Barcelona, April 12th – 14th, 2011
11. P. Kung, L. Wang, M. I. Comanici: Stator End Winding Vibration and Temperature
Rise Monitoring, 2011 Electrical Insulation Conference, Annapolis, Maryland, 5 -
8 June 2011
12. C. Kreischer, H. Steins, S. Strack, J. R. Weidner: New Diagnostic Procedure for
Monitoring of Stator End Winding Vibration at Generators with Rapid Load
Demand Fluctuation, Cigre SC A1 Meeting and Colloquium, Madrid 2015, Spain
13. André Tétreault: On-line End Winding Vibration Measurement Using Fiber Optic
Sensor Technology, EPRI Conference, Barcelona 2011, Spain
14. J. Letal, B. Satmoko, N. Manik, G. Stone: Stator end-winding vibration in two-pole
Machines, Petroleum and Chemical Industry Conference, Cincinnati 2018, USA
15. T. Tanaka, H. Murayama, K. Hattori, K. Takahashi, “Direct measurement of strand
temperature of turbogenerator with FBG sensors” Paper A1-109, Cigre 2012,
Paris, France
16. N. M. Theune, M. Kaufmann, J. Kaiser, M. Willsch, T. Bosselmann, P. Krämmer:
Fiber Bragg Gratings for the Measurement of Direct Copper Temperature of

Stator Coil and Bushing Inside Large Electrical Generators, International
Conference on Optical Fiber Sensors 14th OFS, Venice, 11. - 13. Oct. 2000, Italy
17. T. Bosselmann, E. Abromitis, V. Chernogorski, S. Lindholm, R. Roeding, U.
Schwanengel, S. Strack, M. Villnow, J. R. Weidner, M. Willsch: Design validation of
an air-cooled turbo generator by using bre optic sensors in a shop test, SPIE
Defense, Security, and Sensing, Fiber Optic Sensors and Applications IX,
Baltimore, 26.04.2012, USA
Contact Author: J.R. WEIDNER 

CSE 024
Line Resonance Analysis Implemented in a

Cloud-based Monitoring System for
Energized Power Cables – A Case Study
AUTHORS
B. SONERUD, M. HUSEBY, K. FAGERVIK - Wirescan AS, Norway

N. PAPADANTONAKIS - Total Eren, Greece
Summary
Line Impedance Resonance Analysis (LIRA) is a characterization method akin to
FDS (Frequency Domain Spectroscopy) and TDR (Time Domain Re ectometry), all of
which are commonly used for cable system characterization and diagnostics. These
methods are typically restricted to of ine measurements due to safety requirements
which limits the usability for condition monitoring as this ideally should be
performed continuously and not at discrete intervals. Online condition monitoring is
usually performed with DTS (Distributed Temperature Sensing), a technique based
on temperature measurements via ber optics, but this method is limited to
detecting temperature changes in the system. An online monitoring system based on
LIRA, which can provide detailed data on the condition of the cable system, is
proposed as an alternative and complementary solution to DTS. The proposed
system does not require a galvanic connection to the cables as it uses non-intrusive
signal inducers and sensors; it is easily adaptable to different cable types and
con gurations and provides automated analysis based on continuous
measurements. To ease data management, the proposed system is cloud-based and
uses a data visualization dashboard built upon cable infrastructure maps in which
different cable sections can be highlighted and historical and present data can be
compared. This makes it possible for the user to handle the large data sets. In this
paper the system is described both with respect to hardware and software in
addition to a presentation of a case study implemented at the Kithaironas Wind Park,
installed by Total Eren in Greece. The wind park comprises 30 turbines and 16 km of
medium voltage cables which was installed in 2012. The technical background of the
LIRA methodology is presented as well as the approach towards solving the
additional challenges associated with online measurements such as the galvanic
decoupling of signal injection and pickup.

KEYWORDS
online monitoring - diagnostics - cables - wind farm - line impedance resonance analysis - TDR - FDS
- DTS - asset management
1. Introduction and background

Power cables play an essential role in the power system and the uninterrupted
capability of power transfer is fundamental for the proper operation of any
transmission and distribution network. Despite extensive type testing, factory
acceptance tests, and after installation tests as well as after installation protective
measures such as burial and mechanical protection, cable faults still occur. These
can have several different causes and are usually divided into two groups – internal
causes and external causes. Internal causes include e.g. poor craftsmanship during
installation of accessories, premature ageing or water treeing due to water ingress,
whereas external causes can be damage by anchors or trawling for submarine cables
or by an excavator for terrestrial cables. The recently published CIGRE TB815
Update of service experience of HV underground and submarine cable systems
highlights the different types of faults and their respective prevalence [1]. This
brochure covers high voltage cable systems intended for operation above 60 kV and
considers a period of ten years ending in 2015. 30000 circuit km of land cables and
6000 km of submarine cables was included with a total number of 744 (2.48 per
1000 km/year) faults in land cables and 22 (0.36 per 1000 km/year) faults in
submarine cables reported. 64% of faults have been classi ed as internal and 29%
as external (the remainder as unclassi ed faults). 56% of the faults have been
attributed to the cable itself whereas the rest occurred in accessories. 82% of
external faults are due to third party mechanical damage (e.g. excavators or
anchors). The average outage time has been calculated to 3.5 months making a fault
a costly affair, not only due to the direct repair costs which can be signi cant –
especially for submarine cables, but also with respect to loss of revenue during an
outage. According to [2] the average publicly disclosed repair cost of export cables
for wind farms amount to 12.5 m£. Medium voltage cable systems can be expected to
exhibit different failure statistics than what has been reported in Cigre TB815 as the
design as well as stress levels differ between voltage classes. One example of
medium voltage failure statistics can be found in [3] where the failure statistics of
25% of the South Korean medium voltage network (11000 km of 22.9 kV cables) is
presented as 3.37 per 1000 km/year. This is a higher failure rate than reported for
high voltage cables studied in Cigre TB815.
Due to the high costs associated with faults avoiding them by condition monitoring
and ageing assessment is an attractive feature of any support system. In addition,
being able to rapidly locate a fault to speed up repairs is also desirable. The different
types of faults are commonly divided into shunt fault, open circuit faults and sheath
faults (described in detail in CIGRE TB773 Fault location on land or submarine links
[4]). Fault location is divided into pre-location and pinpointing techniques. Pre-
location can, for instance, be performed with Time Domain Re ectometry (TDR) and
the pinpointing with acoustic methods. Condition monitoring and ageing assessment
can be performed by measuring partial discharge characteristics, voltage withstand
and dielectric response. Line Impedance Resonance Analysis (LIRA) is capable of
both fault detection and condition monitoring during operation [5]–[7].

In this paper a case study of an installation of a line impedance resonance system
modi ed for cable monitoring during operation is presented. The system is
described both with respect to hardware and software as well as implemented as a
case study performed at the Kithaironas Wind Park - installed by Total Eren in
Greece. In addition to online monitoring, the system allows for remote monitoring,
manifested here as the wind farm in Greece being monitored from Norway. The wind
park comprises 30 turbines and 16 km of medium voltage cables which was installed
in 2012. The technical background of the LIRA methodology is presented as well as
the approach towards solving the additional challenges associated with online
measurements such as the galvanic decoupling of signal injection and pickup. The
system is based on [5]–[7] but especially the sensors and the hardware need to be
adapted for non-galvanic measurements. Developments also include cloud-based
data storage and the possibility to access data and perform analysis in an online
dashboard.
2. LIRA methodology
LIRA is based on measurements of the surge impedance of a cable at a large range of
frequencies, thereby assessing the cable condition in the frequency domain. This has
the advantage that the frequency content of the measurement signal can be easily
modi ed according to the most suitable range for the cable that is being measured. A
large range of cable lengths can be conveniently and reliably measured with this
technology – from just a few meters to hundreds of kilometers. The algorithms at the
heart of the LIRA technology are derived from the theory of electromagnetic wave
propagation in transmission lines and the well-known Telegrapher´s Equation.
Given that the measurements are taken in the frequency domain, ef cient frequency
domain algorithms can be applied to produce calculated cable properties, which give
valuable insight into the cable condition. Moreover, given that this technology is fully
transparent regarding the frequency content of the measurement signal, it is
possible to derive these quantities at multiple frequencies, giving valuable insight as
well as quality control over the calculated properties.
In an ideal cable system, the surge impedance is constant. However, discontinuities

such as joints, cable damages and defects cause changes in the impedance at these
discontinuity points, which subsequently cause the incoming signal to be re ected.
This line input impedance signal is then measured at all the applied frequencies,
and effective algorithms can be applied to this measurement signal to determine
characteristics of the cable. A foundational part of LIRA is the analysis of the signal at
the different resonance frequencies that are caused by the properties of the cable,
and LIRA employs processing algorithm for identifying these resonance frequencies,
inclusive of the identi cation of impedance phase zero-crossings. Some details on
using zero crossings for cable diagnostics can be found e.g. in [8].
LIRA calculates physical properties of the cable based on the above, inclusive of
cable capacitance, cable resistance, cable inductance, characteristic impedance,
cable attenuation and other, proprietary property calculations like Delta G, a wide
band tangent Delta calculation. Additionally, the cable impedance changes resulting
from the measurements can be derived and visualized as a function of the cable
length (denoted the Spot Signature of the cable measurement), thereby making it
intuitive and easy to detect and monitor the development of suspected cable fault
locations.

In the online measurement con guration of energized cables, no galvanic contact
with the cable core or screen is possible. This implies that, whilst the LIRA core
technology remains essentially the same as in the measurements of de-energized
cables, the measurement signal must now be injected into the cable through other
means than using a physical contact, and consequently, the re ected signal must be
measured without having physical contact with the conducting materials of the
cable. A key invention here is therefore the sensor design of the LIRA Online
technology.
3. Sensor design
A capacitive and inductive pickup are used to measure the voltage and current
signals induced by the current injector which can be of capacitive or inductive
design. Devising such pickups is a challenge due to several reasons. First, the
injected signal is superimposed on the operating voltage and current as this is an
online technique which means that the power frequency should be ltered out by the
sensors. This implies that an ideal pickup should have a response such as in Figure
1a or 1b with no signal transfer at 50/60 Hz and acting as a band pass lter from 100
kHz to 20 MHz. Second, the gain at the relevant frequency band is limited, mainly
due to the space available in the cable termination bays. The capacitive pickup
consists of a conductive plate which partially covers the surface of the connector. The
larger the plate the more gain which is in turn limited by the available space. This is
also valid for the inductive pickup, manifested as a Rogowski coil in the current
con guration. Third, the frequency response of the actual pickups, shown
schematically in Figure 1c and 1d, differs from the ideal response due to stray
capacitance and coil resistance. For instance, the number of windings, distance
between windings and the loop area are some parameters which in uence the gain
as well as the position and magnitude of the resonance in the Rogowski coil.
4. Data acquisition
An overview of the Data AcQuisition (DAQ) system is provided in Figure 2. This gure
represents the essence of the components that are installed on-site. For each cable
phase studied, inductive and capacitive sensors need to be installed along with the
measurement hardware electronics that controls the signal injection and the system
that receives the signal from the sensors. The output from the measurement
electronics are real and imaginary values of the current and the voltage respectively
at each sampling interval. These values are then passed through a LAN connection
on to a controlling unit (the Gateway), which then packages the signal and passes the
data on to the cloud sub-system. The Gateway is responsible for the management of
the measurement devices and can control multiple such cable measurement
con gurations simultaneously through the local area network. The Gateway is also
responsible for performing data integrity checks and for changing measurement
parameters should this be required.

Figure 1- a. Ideal frequency response of a capacitive pickup / b. Ideal frequency response of an
inductive pickup / c. Actual frequency response of a capacitive pickup /d. Actual frequency response of
an inductive pickup

Figure 2 - Data Acquisition Sub-system
5. Software and dashboard

Some important principles behind the design of the overall system for processing
and presenting the data from the cable under measurement, are that
The system should be extensible and be positioned to take advantage of recent

trends in AI techniques such as Deep Learning, so that signi cant amounts of
data can be analyzed by machines and not only by humans;
The system should be easily scalable to multiple cables and sites;
The system should be results focused and provide a low threshold entry to usage
for both experts and non-experts;
The system should be easy to deploy.
These principles have been guiding the overall design, and a typical Internet of
Things (IoT) architecture has been adopted for the system, where the locally placed
data acquisition system is responsible for collecting the measurements, whereas a
back-end processing and storage sub-system has been implemented as a cloud data
processing solution. The front-end user interface has been implemented as a web-
client so that it can be made available in a web browser on any type of device, be it
mobile phones, tablets, or desktop.
The overall system is illustrated in Figure 3. The DAQ sub-system communicates

with a data ingestion Application Programming Interface (API), which also has the
responsibility of distributing the data to the appropriate micro-services for
processing. The data processing sub-system is an extensible system that can be
modi ed according to the needs of the processing and is currently implementing

standard processing according to the LIRA methodology [5]–[7]. It is envisaged that
deep learning algorithms and analysis functions will be deployed as part of the
work ow in the future.
Given the nature of the data in that new measurements are repeatedly captured and
analyzed, it has been an important aspect to capture the data contextual information
in the presentation layer/dashboard, as and when new measurements are available.
Also, the dashboard functionality should be highly visual, and should provide the
essential information for each measurement at a glance. The main element of the
dashboard has therefore chosen to be the cable under measurements, displayed in a
map, so that it is easily observed which cable is measured, and what the data is.
Figure 3 - The Analysis and Presentation Sub-system
6. Case study: Total Eren

Total Eren is an international player in the renewables energy generation industry,
generating over 3200 MW of renewable power, and has a further 2200 MW under
construction. Total Eren operates solar plants, wind farms and hydroelectric plants in
Latin America, Africa, Central and South Asia, Asia Paci c, and Europe.
6.1. Overview of Kithaironas wind farm

The Kithaironas wind farm is a mature wind park operated by Total Eren in Greece.
The wind park is in the Viotia region, about 1.5-hour drive northwest of Athens,
where the 30 G52-850kW wind turbines generate around 86.5 GWh per year. The
park is situated on a mountain ridge approx. 1000 m above sea level, as illustrated
in Figure 4.

Figure 4 - Kithaironas Wind Farm Location and con guration

During this time, several issues have
been observed with different cables in
the park, especially with regards to
export cables. Total Eren’s interest in
online Line Impedance Resonance
Analysis arises from the need of
mitigating down-time periods and
repair costs through continuous cable
health monitoring, both speci cally for
the Kithaironas Wind Farm and
strategically for the company.
The wind farm consists of two daisy

chained wind turbine strings, connected
to a substation which subsequently
feeds power to the grid. Each string
connects 15 wind turbines to the
substation through two export cables.
The cables, AL/XLPE/CWS/PVC
1X630/16 RM, 12/10(24) kV, are
manufactured by Hellenic Cables S.A. in
Greece. Each cable consists of three
single phases laid together in a trench
from the rst turbine on the string to the
substation. The cable lengths are
approx. 2600 and 5000 meters.
Each phase is terminated at the

turbines switch gear by means of T-type Figure 5 - Measurement equipment installed on
slip on connectors for C interface. The T-Connector
connectors are manufactured by TE
Connectivity and are of type RSTI-5953.
These connectors have a relatively large
surface which is bene cial for size and
placement of the capacitive sensors.
Also, given the size of the connectors,
suf cient space is available to
comfortably t the electronics bracket
and the inductive pickup without any
room/space constraints. The installation
of the sensors and the hardware
electronics on this T-type connector is
depicted in Figure 5.
6.2. Implementation
The collaboration between Total Eren and Wirescan was initiated in 2019 with
technical studies of cables, terminations, and practicalities around constructing an
IoT system for cable monitoring on this speci c windfarm. The system was
eventually installed, commissioned, and put into operation in November 2020 [1].

The systems for both cables were installed during one working day, and once the
installation was complete the sensors started to gather impedance information from
the live cables. The cloud connection was set up automatically by the on-site DAQ
system, which will also secure the reestablishment of the connection even if a power
outage or loss of internet connection should occur.
6.3. Operation
The system has at the time of writing been in operations for 75 days. The system has
behaved as predicted, although some external events have allowed for appropriate
stress testing of the system, including power failures and loss of internet. However,
the system has re-established the measurement campaign and the connectivity to
the cloud system every time such an event has occurred. The dashboard for
monitoring the activity for one of the cables under measurement is shown in Figure
6.
Figure 6 - Dashboard for monitoring the real-time measurements remotely
7. Discussion and conclusions

Ageing assessment and condition monitoring during operation of power cables is an
attractive feature and has the potential of substantial savings through the
prevention of outages and rapid fault location. This is especially true if the data and
analysis is made accessible via online services and can be continuously accessed
from a remote location. The case study highlighted here shows the potential of such a
system as well as the added value of cloud usage with regards to collection,
synthesis, and visualization of data from LIRA measurements. As any good case
study, it highlights areas of improvement. Further developments include
optimization of sensors and hardware with regards to noise sensitivity as well as
development of con gurations for different types of end connectors. There is also
room for improvement with regards to long term stability of sensor gain as well as
the data collection robustness.
References

1. CIGRE Technical Brochure 815, “Update of service experience of HV
underground and submarine cable systems,” 2020.
2. Offshore Wind Programme Board, “Export Cable Reliability: Description of
Concern,” vol. 2017, no. July, 2017.
3. M. Bawart, T. Neier, K. Jens, and S.-M. Kim, “Estimation of the Remaining Life
Time of Medium Voltage Power Cables based on the Aging Index using Unique
Algorithms and a Large Database,” in 2020 IEEE/PES Transmission and
Distribution Conference and Exposition (T&D), Oct. 2020, pp. 1–5, doi:
10.1109/TD39804.2020.9299991.
4. CIGRE Technical Brochure 773, “Fault location on land and submarine links (AC &
DC),” 2019.
5. P. F. Fantoni, “Condition Monitoring of Electrical Cables Using Line Resonance
Analysis (LIRA),” in Volume 1: Plant Operations, Maintenance, Engineering,
Modi cations and Life Cycle; Component Reliability and Materials Issues; Next
Generation Systems, Jan. 2009, pp. 171–178, doi: 10.1115/ICONE17-75833.
6. G. J. Toman and P. F. Fantoni, “Cable Aging Assessment and Condition Monitoring
Using Line Resonance Analysis (LIRA),” in Volume 1: Plant Operations,
Maintenance, Installations and Life Cycle; Component Reliability and Materials
Issues; Advanced Applications of Nuclear Technology; Codes, Standards,
Licensing and Regulatory Issues, Jan. 2008, pp. 177–186, doi: 10.1115/ICONE16-
48523.
7. P. F. Fantoni and Anders Nordlund, Wire System Aging Assessment and Condition
Monitoring ( WASCO ). 2006.
8. D. Rogovin and R. Lofaro, Evaluation of the Broadband Impedance Spectroscopy
Prognostic / Diagnostic Technique for Electric Cables Used in Nuclear Power
Plants. Washington: U.S. Nuclear Regulatory Commission Of ce of Nuclear
Regulatory Research Washington, 2006.
Contact Author: B. SONERUD 
[1]
The installation was signi cantly delayed due to the on-going COVID-19 pandemic

CSE 024
Comparison of electrical clearance

between Japan and other countries
AUTHORS
T. YAMANAKA - Electric Power Development Co., Ltd., Japan

Y. SHIMOMURA - Kyushu Electric Power Transmission and Distribution Co., Inc.,
Japan (Former af liation: The Japan Electric Association)
S. MURAKAMI - J-POWER Transmission Network Co., Ltd., Japan
K. FUJITA - Tohoku Electric Power Network Co., Inc., Japan
Y. KOYAMA - Chubu Electric Power Grid Co., Inc., Japan
Summary
The electrical clearance (D) in Japan is calculated from the sum of Safety distance
(Ds) and Minimum air clearance required to prevent a disruptive discharge between
phase conductors and objects at earth potential during fast front or slow front
overvoltages (Del). Del is ‘clearance increment’ and depends on the system voltage.
This is the same idea as in foreign countries. The interpretation of the technical
standards (Ministerial Ordinance) for electrical appliances and materials (hereinafter
referred to as ‘Interpretation of Technical Standards’) stipulates a clearance value
that increases with rising voltage (hereinafter referred to as ‘clearance increment’).
The clearance increment varies depending on the safety distance and has been set
as 12 cm / 10 kV above 60 kV for other structures and trees, and 15 cm / 10 kV above
35 kV for buildings.
The clearance increments in other countries are generally set to about 6 cm / 10 kV,
so as the voltage rises, Japan's clearance is clearly larger than in other countries.
When the clearance was established in Japan in 1959, there were no 500 kV
overhead transmission lines (hereinafter referred to as ‘OHL’), and the clearance of
extra-high voltage OHL had not been suf ciently studied. Since then, the number of
OHL of 170 kV class and above has been increasing rapidly. It was thought that if
these clearances could be reviewed to be comparable to those in other countries, the
height of towers would be reduced and the area of tree trimming that interferes with
OHL would be reduced, leading to cost savings in the construction and maintenance
of OHL. Based on these circumstances, in order to explore the possibility of
reviewing the electrical clearance of EHV OHL in Japan, we conducted overseas
survey on the regulations of Germany, France, Canada, and the United States and
then not only the electrical clearance values but also basis for their setting were

investigated and compared to Japanese one. The overseas survey con rmed that
there would be no safety problem with relaxing the electrical clearance of OHL
exceeding 170 kV, hence we applied to the regulatory authority to change it. As a
result, on February 25, 2020, it was approved to amend the clearance increment
(Del) to 6 cm / 10 kV with respect to the electrical clearance between overhead
transmission line exceeding 170 kV and building, road, other electric wire, tree.
This amendment of the electrical clearance will reduce the height of 500 kV towers
by about 2 m, thereby reducing the cost of materials, transportation, and assembly of
the towers. In addition, by reviewing the electrical clearance from the conductor to
the tree, the area of trees to be cut down as obstacles can be reduced, especially
when the conductor swings sideways in the wind. Therefore, it will reduce not only
the compensation cost of new OHL, but also the maintenance cost of existing OHL.
This paper describes the results of the studies and investigations that formed the
basis for the rst revision of Japan's electrical clearances in 61 years, and the effects
of the regulatory changes on the design and cost of OHL.
KEYWORDS
Overhead transmission line - Clearance - Minimum insulation gap - Interpretation of technical

standards for electrical facility in Japan - National Electrical Safety Code (NESC) - CSA - EN 50341-1 -
EN 50341-2-4 - EN 50341-2-8 - VDE - UTE C11
1. Introduction
The interpretation of the technical standards (Ministerial Ordinance) for electrical
appliances and materials (hereinafter referred to as ‘Interpretation of Technical
Standards’) stipulates a clearance value that increases with rising voltage
(hereinafter referred to as ‘clearance increment’). The clearance increments vary
depending on the type of object approaching the conductor, with ‘other
structures*1 and trees’ set at 12 cm / 10 kV and ‘buildings’ at 15 cm / 10 kV. However,
in other countries, they are generally set at around 6 cm / 10 kV, and as the voltage
class increases, the deviation from the Japanese clearance becomes more
pronounced.
This standard value was set in 1959, but at that time, 500 kV overhead lines did not
exist and the clearance for EHV lines was not fully considered. Since 1959, the
number of EHV OHL above 170 kV has been increasing rapidly. Therefore, if this
clearance can be revised, the cost of construction and maintenance of OHL will be
reduced.
*1In Interpretation of Technical Standards, structures other than ’buildings, roads (except those
with infrequent vehicular and human traf c), pedestrian bridges, railroads, tracks, cableways,
overhead communication lines, low-voltage (600 V or less) or high-voltage (7,000 V or less)
overhead electric lines, low-voltage or high-voltage train lines, and other electric lines
exceeding 7,000 V’ are de ned as ‘other structures’.
2. Comparison of clearance distance in EHV

overhead transmission lines

The clearance (D) is calculated from the sum of Safety distance (Ds) and Minimum air
clearance required to prevent a disruptive discharge between phase conductors and
objects at earth potential during fast front or slow front overvoltages (Del). Del is
‘clearance increment’ and depends on the system voltage. (Table 1).
The safety distance (Ds) was calculated from model cases in each country, and there
was no signi cant difference between Japan and other countries for buildings and
other structures. On the other hand, the clearance increment (Del) is 6.4 cm / 10 kV
in the United States and Canada, 6.2 cm / 10 kV in Germany (average of 35 kV or
more), and 7.5 cm / 10 kV in France, so the clearance increment in Japan is
remarkably large. As a result of this, the divergence from Japan's clearance becomes
striking as the voltage class grows larger (Figure 1 and 2).
Table 1 - Calculation formula for the clearance (D) and References
Calculation formula for the

Country clearance (D) Reference
United States D*2 = Ds + Del - National Electrical Safety Code

(NESC)
Canada D*2 = Ds + Del - CSA standard C22.3 No.1-15

Overhead Systems
Germany D = Ds + Del - EN 50341-1, EN 50341-2-4

(VDE 0210-1, VDE 0210-2-4)
France D = Ds + Del*3 - EN 50341-1, EN 50341-2-8

(UTE C11-001, ARRÊTÉ
TECHNIQUE)
Japan D = Ds + Del*4 - Technical standards for electrical

facility in Japan
*2 Phase to ground voltage

*3 Consider the probability of approaching the conductor
*4 This value varies depending on the object that must be kept away from the conductor
3. The clearance increment (DEL)
3.1. Clearance increments in Japan (old standard)

The clearance increment in Japanese old standard was set to increase 15 cm per
every 10 kV exceeding 35 kV for buildings, and 12 cm per every 10 kV exceeding 60
kV for other structures and trees [1]. This value was set by the Electric Construction
Standards of 1959. According to the description in the standard, the basis of the
clearance increment of 12 cm / 10 kV is as follows. It had to have a safety margin
slightly greater than the Minimum insulation gap*5 of 6 cm / 10 kV and the Standard

insulation gap*6 of 10 cm / 10 kV and also refer the other countries values. Therefore
the value was decided to adopt the intermediate value between those up to 1958 (20
cm / cm) and the Minimum insulation gap (Figure 3). Furthermore, since it was
considered that the value for buildings needed a safety margin more than other
structures, 15 cm / 10 kV was adopted.
*5 ‘The minimum insulation gap’ is de ned as the air gap that does not cause ashover for the
assumed switching surge and short-time overvoltage.
*6 ‘The standard insulation gap’ is de ned as the air gap that allows lightning-induced ashover
to occur between the arc horns and not between the wires and the tower.
Figure 1*7: Comparison of electrical clearance Figure 2*7: Comparison of electrical clearance
with other countries (other structures) with other countries (buildings)
*7Figures 1 and 2 are taken from Figures 3.1 and 3.3 in Electrical clearance for extra high
voltage overhead transmission lines exceeding 170 kV, JESC E2012 (2013), (in Japanese) [13]
3.2. Clearance increment in the U.S.

The current U.S. clearance increment is set to 1.0 cm / kV above the ground voltage
of 22 kV [2]. In 1949, it was to 0.5 inch (1.27 cm) / kV, but in the 1980s it was
amended to a 0.4 inch (1.0 cm) / kV [3]. The 0.5 inch and 0.4 inch per 1 kV of ground
voltage can be calculated from the Guard Zone speci ed in the NESC. Guard Zone
means the Minimum insulation gap to the charged section (Figure 4). The Guard
Zone has been updated with the advancing research and technology on switching
surges since around 1960, and presumably as a result, the clearance increment was
amended from 0.5 inch / kV to 0.4 inch / kV. In addition, considering the switching
surge factor used in EHV OHL, the clearance increment exceeding 35 kV is 0.64 cm /
kV.
3.3. Clearance increment in Canada

The clearance increment in Canada is set at 1.0 cm / kV when the ground voltage
exceeds 22 kV, and is equivalent to the United States clearance increment [4]. The
CSA standard adopted in Canada cites a number of IEEE standards, and the
clearance increment has presumably been set in the same way as the US NESC
standard.

Figure 3 - Clearance increment and insulation
Figure 4 - Clearance from live parts
gap that change with increasing system voltage
3.4. Clearance increment in Germany

The German clearance increment is based on the European reference EN standards
[5][6][7]. This value is speci ed as the Minimum insulation gap based on the
European empirical method, and was determined by analyzing values from about 30
countries that cooperated in establishing the standard (Table 2).
Table 2 - Minimum air clearance Del
Highest system voltage

Us (kV) Del (m)
36 0.35
52 0.60
72.5 0.70
100 0.90
123 1.00
145 1.20
170 1.30
245 1.70
300 2.10
420 2.80
525 3.50

3.5. Clearance increment in France
The clearance increment in France considers the probability of approaching the
conductor. The adopted value depends on the circumstances, so if the probability of
approaching the conductor is low, there is an increase of 0.25 cm / kV, for medium
probability the increase is 0.50 cm / kV, and for high probability the increase is 0.75
cm / kV [8][9].
4. Minimum isolation gap in Japan

In 1958, Japan's rst proposed recommendation for the insulation design standard
was drafted. This proposed recommendation was drafted with reference to the
technical documents and test data of other countries, and the Minimum insulation
gap is set as the clearance to withstand switching surges and sustained abnormal
voltages [10]. Later, based on advances in research and technology on switching
surges, insulation design guidelines were published in 1966 and 1986 [11] [12].
When the clearance increment is calculated from Minimum insulation gap against
switching impulse voltage in Japan, it is 5.2 cm / 10 kV (Figure 5). A more
conservative value would be 6.0 cm / 10 kV, which is equivalent to the clearance
increment in other countries.
Figure 5 - Calculation the clearance increment from Minimum insulation gap against switching impulse
voltage in Japan
5. Comparison of electrical clearance between

new and old standards
Comparing the clearance increment before and after the amendment of the
interpretation of the Japanese technical standards, the clearances could be reduced
by about 2 to 3 m for 500 kV (Figure 6 and 7). As a result, tower height of 500 kV OHL
can be reduced by about 2 m.

Figure 6 - Clearance distance before and after amendment (other structures)
Figure 7 - Clearance distance before and after amendment (buildings)
6. Potential for flashovers resulting from the

amendement of the clearance increment
The possibility of ashover from the conductor to other objects after the clearance
amendment was examined as follows, in the case of lightning striking the conductor
(lower phase). The standard insulation gap is an airgap set to ensure that ashovers
caused by lightning occur between arcing horns rather than between conductor and
tower, and it is calculated by the following equation [12].
L=1.115 x Z＋0.021
Where,
L: Standard insulation gap (m), Z: Arcing horn gap (m)

In the new standard, a clearance exceeding the standard insulation gap has been
assured, and there is no potential for ashovers to other objects (Table 3, Figure 8
and 9).
Table 3 - Standard insulation gap and clearance
System voltage Standard Clearance to objects (m)

(kV) insulation
gap (m) [10] New standard Old standard
Other Buildings Other Buildings

structures 6 cm / 10 kV structures 15 cm / 10 kV
6 cm / 10 kV 12 cm / 10 kV
187 1.53 (100%) 3.44 (225%) 5.22 (341%) 3.56 (233%) 5.40 (353%)
220 2.36 (100%) 3.62 (153%) 5.40 (229%) 3.92 (166%) 5.85 (248%)
275 3.03 (100%) 3.98 (131%) 5.76 (190%) 4.64 (153%) 6.60 (218%)
500 4.82 (100%) 5.30 (110%) 7.08 (147%) 7.28 (151%) 10.05 (209%)
Figure 8 - Comparison of the clearance (between

Figure 9 - Comparison of the clearance (between
conductor and other structures, trees)
conductor and buildings)
7. The effect of regulatory amendment in

electrical clearances on the design and cost of
overhead transmission lines
7.1. Estimated construction cost reduction for overhead

transmission lines

By amending the electrical clearance between the EHV OHL and the trees, the
clearance between the conductor of the 500 kV OHL in the mountainous area and the
trees can be reduced by 1.98 m. Therefore, the height of the tower can be reduced by
about 2 m (See Figure 10). The reduction of the tower height by 2 m means that
approximately 2 tons of steel per tower can be reduced, and the cost of tower
materials and erection work can also be reduced. This amendment will contribute to
a reduction of about 1 % of the total construction cost of the 500 kV OHL in Japan
[13]. Since most of the EHV OHL in Japan pass through mountainous areas, the
economic effect of the revised clearance regulation is incredibly signi cant. The
electrical clearance of OHL in urban areas (the minimum height above ground of
OHL)is determined by the regulation value of electric eld strength at the ground
surface. It has been con rmed through actual measurements that the EMF level is
not a problem if the OHL satis es this electric eld strength regulation value.
Figure 10 - Reduction effect of EHV overhead transmission line towers passing through mountainous
areas
7.2. Effect of reducing the area of trees that are cut down
as obstacles to conductor
The expected reduction in the area of tree falling on conductors due to the
amendment of electrical clearance is described as follows. The reduction ratio in the
area of tree trimming, determined from the required clearance when the conductor
swings 60 degrees and sideways in high winds, is 3.8 % for overhead transmission
lines of 170 kV and above, which can be reduce the compensation cost (see Figure
11) [13]. This is not a temporary compensation at the time of OHL construction, but
lasts as long as the facility exists, so this economic effect is also signi cant.

Figure 11 - Comparison of falling width of obstructed trees according to new and old standards
8. Conclusions
As in Japan, the electrical clearance in other countries is calculated based on the
standard clearance value plus the clearance increment per 10 kV, but since the
Japanese clearance increment used to be larger than in other countries, there was a
maximum divergence of about 3 m in 500 kV. The Japanese electrical clearance was
amended for the rst time in 61 years on February 25, 2020 after many years of
study and recent overseas surveys. This will reduce the height of 500 kV towers by
about 2 m, which will reduce the material and assembly costs of the towers and
reduce the area of trees to be cut down as obstacles. It will not only reduce the cost of
constructing new OHL, but also the cost of maintaining the existing one, which will
have a signi cant economic impact.
References
1. Technical Standards for Electrical Facility in Japan (in Japanese)
2. National Electrical Safety Code (NESC) C2 (2017)
3. National Bureau of Standards Handbook H43
4. CSA standard C22.3 No.1-15 Overhead Systems (CSA)
5. EN 50341-1 Overhead electrical lines exceeding AC 1 kV - Part 1: General
requirements - Common speci cations
6. EN 50341-2-4 Overhead electrical lines exceeding AC 1 kV - Part 2-4: National
Normative Aspects (NNA) for GERMANY
7. VDE 0210-1 Overhead electrical lines exceeding AC 1 kV Part 1: General
requirements – Common speci cations
8. EN 50341-2-8 Overhead electrical lines exceeding AC 1 kV - Part 2-8: National
Normative Aspects (NNA) for France
9. UTE C11-001 Conditions techniques auxquelles doivent satisfaire les
distributions d'énergie électrique - Arrêté technique
10. Insulation Design Standard for Power Transmission Lines, IEEJ Recommendation,
1958 (in Japanese)

11. Insulation Design Guidelines for Overhead Transmission Lines (Part I), IEEJ
Technical Report, 1966 (in Japanese)
12. Insulation Design Guidelines for Overhead Transmission Lines (Part II), IEEJ
Technical Report, 1986 (in Japanese)
13. Electrical clearance for extra high voltage overhead transmission lines exceeding
170 kV, JESC E2012 (2013), (in Japanese)
Contact Author: T. YAMANAKA 

CSE 024
Eco-design as a systemic innovation

method for offshore substations
AUTHORS
D. BOTRUGNO, M. NUNES - RTE
Summary
The energy transition and the ght against climate change are among the crucial
challenges of current times and a major focus of RTE's commitment. As a TSO, RTE
takes actions to reduce the environmental footprint of its new projects, enhance the
exibility and resilience of the transmission network and facilitate the integration of
renewable energies.
RTE believes new approaches are essential in this context in order to innovate and
nd sustainable solutions to overcome these challenges.
In this light, RTE is extending its efforts through an ambitious and comprehensive
eco-design approach. RTE is raising awareness among its teams and disseminating
the necessary tools so that taking the environment into account in each decision,
along with the technical and economic criteria, becomes engrained in the company’s
practices. In 2020, RTE carried out the LCA (Life Cycle Assessment) of an electrical
offshore substation, with an expert rm in the eld.
This article’s goal is to share the main lessons, both at the methodological and
results level, with the CIGRE members. The identi cation of the main sources of
impact and related ways of implementing eco-design are part of the study's
deliverables and should enable improvements to be taken since the short term (and
along the assets life).
The LCA methodology is based on a multi-step and multi-criteria analysis, from the
extraction of raw materials to the end of infrastructure life. The perimeter of the LCA
is the full offshore substation (all electrical equipment, auxiliary equipment,
platform systems and structures) all over the asset life, except the HV export cables
and the IAC cables.

The process is iterative and, over the time, allows to build a quantitative model
where the characteristics of each component are identi ed, the environmental
impacts are calculated, all over the asset life. Several alternatives in design have
been tested in the model, covering driving choices for the equipment and the
infrastructure. The results have enabled RTE to understand the factors that allow to
reduce the environmental impact and make the offshore substation as sustainable as
possible.
Finally, the future step of implementation are described.
The main ambition is to drive the company choices in design, supply chain,
construction, operations and maintenance aspects. The vision is to reduce the
environmental footprint of all the infrastructure all over its life cycle, create a
positive integration within the environment, foster technical and social innovations,
and support new solutions for the sustainable development of a future decarbonized
system infrastructure.
KEYWORDS
Offshore – Substation – LCA – Eco-design – Innovation – Environment – Sustainability
1. Introduction
As part of the energy transition and more speci cally of the offshore renewable
energies development, RTE, the French TSO, has extended its activities to the
maritime domain and included within its scope the offshore substations for the
future French offshore wind-farms (from public tender AO3 awarded in 2019
onwards). In the context of offshore grid connection projects, RTE plays a major role
in the planning and development of offshore renewable energies, both at a pilot and
commercial scale. According to the multi-year energy programme (French PPE, [1]),
the objective of wind power at sea is to achieve a power of approximately 5 GW in
2028. RTE is responsible for nancing, designing, building and operating the
offshore power infrastructure, its substations (platforms) and HV cables network.

Figure 1 - overview of the main French offshore grid connection projects (as of summer 2021)
RTE is committed to reducing its environmental footprint, and is recognized as such,

in particular by stakeholders and auditors in charge of ISO14001 certi cation
renewal. In order to be more ef cient, the company has decided to systematically
rely on eco-design, deployed throughout all the company processes.
The central objective is to reduce RTE's global environmental footprint in an
effective way, thanks to a systemic vision of its impacts over the entire life cycle of
each asset and activity. This vision also offers a new leverage in favour of technical-
economical optimisation, environmental and societal innovation, and a potential
reduction of risks on the value chain.
The process covers the entire scope of RTE’s activities, meaning power infrastructure
and assets development, management of the electrical system, and all support
services.
This ambition is accompanied by the upstream engagement of a constructive

dialogue with stakeholders on these issues, and in particular with RTE suppliers.

Finally, the company is at the heart of an energy transition that faces great
uncertainties. Solutions must therefore mobilize the minimum of inputs through the
research of focus and sobriety (analysis of the needs), optimize the ef ciency, and
minimize the impacts of any resulting strategy modi cation.
This aim is pursued by seeking modularity and reuse / recyclability, by developing

an organic approach and methodology at the level of both new power infrastructures
to be built and existing to be renewed, and by integrating the context of the evolution
of the energy system as a whole.
Among the axes of development of eco-design as a systemic innovation approach,

the realization of several LCA studies. LCA’s allow RTE to assess the environmental
footprint of the structuring components of the future power infrastructure network,
to identify the main impact factors, to evaluate the possible alternative and make
considered choices in order to improve their environmental impact over the life of
each asset.
Some of the main environmental factors assessed through these LCA’s are: the
impact on reduction of greenhouse effect gases, the depletion of natural resources,
eutrophication and acidi cation of seas, and air pollution. These indicators go back
into the iteration of design choices, proposing ways and strategies for the eco-design
improvement and consolidation into the company processes and assets.
2. The urge of turning environmental challenges

into opportunities
Involved since more than a decade into the energy transition, the major players
share the responsibility of innovating and preparing the power infrastructure
network to implement the EU Green Deal.
Eco-design innovations are crucial for the transition to a sustainable, regenerative,
inclusive and circular system. These initiatives can create a wide range of results,
improving solutions for a sustainable economy ( ghting against climate change and
environmental degradation) and fostering the protection of environment (limiting as
much as possible negative impacts and creating positive impacts on eco-systems
regeneration and biodiversity). They can also support better responses of the system
to global matters (including pandemic crisis, and resulting disruption of supply
chain).
However, designing an asset in such a way to integrate the environment as one of the
key drivers needs to cover all the asset life cycle. Within this context and with the
ambitions described in the previous chapter, RTE decided then to perform a life cycle
assessment (LCA) for a full offshore substation scope.
Although the LCA method is well-known and used by the industry, our
bibliographical study put in evidence that the present work study represents a rst-
in-its-kind (at the time of writing and as regards to the study perimeter and its
approach).
The study combines: a comprehensive perimeter of a full offshore substation (all

infrastructure, all equipment, all phases of its life); its iteration to nd alternative
choices and its systemic approach to design (thus, feeding back the choices and
being integrated as a standard practice for all asset strategies).

3. Offshore substation life cycle assessment
methodology
The proposed approach is based on the application of the Life Cycle Assessment
(LCA) methodology.
The LCA is an internationally standardized method (ISO 14040 to 14044) which

allows to assess the quanti able effects on the environment of a service or product
from the extraction of the necessary materials, from its development to end-of-life
channels. This is a so-called multi-step method.
The method consists in carrying out an exhaustive assessment of the consumption of

natural resources; energy consumption and environmental emissions (releases to
air, water, soil and waste) from the entire system studied.
A rst step is to draw-up an exhaustive inventory of the inputs-outputs speci c to

each step of the system. These reports of incoming and outgoing ows are called Life
Cycle Inventories (LCI). Each parameter has been calculated based on an offshore
substation preliminary basic design developed by RTE. To model the platform, RTE
chose an advanced modelling software. The platform model can be easily adapted
with new parameters to test new ideas or new platform types. Furthermore, the
software includes large environmental databases for each raw material and product.
These databases are the link between the data collection and the environmental
metrics.
The ows of materials and energy taken and released into the environment at each
stage are then weighted and aggregated to quantify environmental impact
indicators. For example, the indicator of climate change, expressed in kilograms of
CO2 equivalent, results from the sum of greenhouse effect gas emissions weighted by
the global warming factors of the various gases.
LCA is thus a multi-criteria methodology: indeed, there is no single environmental

score, but the results of the study are presented in the form of several environmental
impact indicators.
Given the innovative nature of the study, an iterative process has been put in place
to perform the analysis [2]. In the LCA context, an iteration is an action loop making
it possible to collect the input data, to model the system in the LCA software, to
output the results of the analysis and to interpret them.
Two iterations were carried out:
The rst iteration allowed to de ne the structuring methodological choices for

the realization of LCA (e.g. system boundaries, functional unit, indicators). These
rst results led to the identi cation of the factors to re ne the conclusions (e.g.
additional data, de nition of scenarios, sensitivity) as well as to start the
identi cation of eco-design suggestions of improvement / development;
Several alternatives in design have been tested in the model, covering driving
choices for the equipment (e.g. technology / type, materials, place of fabrication,
means of transportation) and the infrastructure (e.g. technology / type, materials,
place of fabrication, means of transportation) – but also regarding the site

conditions (e.g. water depth, nature of soil and scouring properties). This second
iteration allowed to freeze the results and their interpretation.
Generally, the main functions of an offshore wind AC substation include:
Collecting the power produced by the wind turbines generators;

Stabilizing it and raising the voltage (e.g. from 66 kV to 225 kV);
Ensuring the electrical protection of the inter-arrays;
Enabling 225 kV transmission to the onshore network, minimizing the transport
electrical losses and reducing the number of subsea cables required.
The typical scheme of French offshore grid connections and the position of the OSS
are shown in Fig. 2 below.
Figure 2 - scheme of French offshore grid connection (and OSS)
An offshore electrical substation is generally composed as follows:
The topside (usually 3 to 5 oors) which houses the HV/LV electrical equipment,
the control and protection equipment, the auxiliary systems, the HV subsea
cables connections and the other platforms facilities
The substructure, which supports the topside. Different types of substructures
are possible (monopile, jacket or concrete). The substructure also support the j-
tubes for the routing and mechanical protection of subsea cables from
underwater to on-board the topside.
The topside and substructure assembly is xed to the ground by foundations, the
nature of which varies in particular depending on the soil characteristics, the
topside size and weight and the meteocean conditions (varying from hollow metal
piles driven into the ground to drilled piles, pre-piled or post-piled structures).
As a reference and for entry data to the study, a preliminary basic design develop by
RTE was taken. For the electric power equipment, the information provided was more
detailed as it was sourced from RTE’s equipment panel.
For the study, the following assumptions are considered for the offshore substation
design:

600 MW power capacity
66 kV / 225 kV voltage levels
normally unmanned platform
steel structure and substructure
The LCA perimeter does not include the 225 kV subsea cables. The reason of such
choice is that the LCA objective is to isolate and analyse the impacts of the offshore
substation from an eco-design perspective. For the same reason, the 66 kV inter-
array cables and the rest of offshore wind-farm are also outside the scope of the
study, this being under the responsibility of the offshore wind-farm developer
company.
In view of the bibliographic analysis carried out as part of this study, the Functional
Unit (FU) retained for this study is: "Collect the offshore wind-farm power production
and raise voltage level from 66 kV to 225 kV, for an installed capacity of the wind
farm of 1 MW and a water depth of 18 m, and for a design life of 25 years".
Finally, more than 98% (by mass) of the elements were modelled and the Circularity
Footprint Formula called CFF method (European Commission, DG ENV) was used to
model the end of life.
This new method, aimed at evaluating the contribution of products to the circular
economy, was developed to modify the rules for allocating the impacts / bene ts of
recycling as well as to take into account additional factors making it possible to
integrate degradation of materials during recycling.
The LCA study and methodology has made the object of a 3-months “critical review”
by an independent panel, according to the requirements speci ed in the ISO14067
standard in the case of possible communication to the general public. A panel of
technical experts and stakeholders was constituted, with the aim to challenge and
corroborate the eco-design hypotheses, data and suggestions for improvements, and
to allow the communication of results to the general public.
4. Life cycle assessment results

Below are presented and discussed some of the main outcomes from the LCA study.
The results are given per functional unit and with clear indication of the phase
originating the impact, either in the construction phase (fabrication, transportation,
installation) or in the service phase (operation and maintenance, deconstruction,
recycling / disposal).
4.1. LCA Results - rst iteration

Figure 3 - Global LCA results for the OSS (per functional unit: 1 MW)
Figure 3 diagram shows the global results from the rst iteration (base calculation)
LCA on an OSS entire cycle. Impacts are shown on a series of indicators considered in
the study:
Climate change
Particulate matter
Photochemical ozone formation
Acidi cation
Freshwater eutrophication
Abiotic depletion (elementary, reserve base)
Cumulated energy
For each indicator (column), the cumulative max / min values are reported.
The results are differentiated per phase of the OSS life cycle (in different colours). It
appears that, for the entirety of the considered indicators, the phases bringing the
most signi cant impacts are the Manufacturing (fabrication) and Operations (O&M)
phases.
It is worth observing that the impacts appearing in green in the diagram correspond
to negative values (thus indicating impacts which are to be avoided). This is the case
of recycled material (used for fabrication instead of new raw materials,) or to the
recycling of these materials at the end of asset life (instead of waste disposal), as
recorded in the CFF method. On the opposite, the impacts appearing in positive
values in the graphs represents impacts onto the environment.
Focusing on the Manufacturing (fabrication) phase, three categories contribute with

the most signi cant impacts (ref. Fig 4):
The steel structure

The electric power equipment, mainly:
Main Power Transformers
GIS (Gas Insulated Switchgears)
NEAT (Neutral Earthing Auxiliary Transformers)
The C&P system and Telecom equipment

Figure 4 - Results from the manufacturing phase
Steel is of course the main substation structure component. Our results show that
steel production, which generates high GHG emissions (4 ton CO2e/ton of primary
steel), is responsible for the majority of impacts resulting from the structure.
Electrical equipment production has a signi cant impact on the reduction of abiotic
mineral resources, in particular because of the production of copper contained in
transformers and lead contained in batteries.
Manufacturing of C&P equipment and Telecom systems has the greatest contribution
to the eutrophication potential of freshwater, due to the manufacture of the electronic
components constituting this equipment. Electronic equipment is made up of several
metals, including metals that are more or less critical from a geological point of view.
The more critical the metal, the more the extraction and re ning stages will generate
energy consumption and local impacts on natural environments, and especially on
freshwater life.
Focusing on the structure fabrication analysis (ref. Fig. 5 below), the piles (in blue)
and the jacket (primary structure in red, secondary in grey) have the highest impacts
on most of environmental indicators (from climate change to cumulated energy)
apart from the decrease in abiotic resources, which is mainly impacted by the
painting and corrosion protection systems (for the zinc / aluminium coatings
fabrication).
Figure 5 - Results from the structure manufacturing

Within the manufacture of electric power equipment (Fig. 6 below), the Power
Transformers and the GIS show the highest impact on all the indicators. For example,
Power Transformers represent almost 50% of the impact on climate change and 55%
of the impact on acidi cation and GIS represent 30% on climate change and more
than 20% on acidi cation. This is in particular due to the large quantity of metals
that compose them.
Figure 6 - Results from the electric power equipment manufacturing
Focusing on the Operations (O&M) phase, the impacts of this phase are strongly
dominated by the electrical losses of the power transmission asset. With reference to
the gure 3) “Results of the global LCA (per functional unit: 1MW)”, those are
represented by the yellow contributions (identi ed as “Use” in the diagram).
Interestingly, it clearly appears the signi cant contribution of the operations and
maintenance life of the asset compared to the construction and installation phases;
as a consequence, the positive changes that can be brought on the electrical
transmission performance acquire importance in the light of the global LCA.
It is important to note that the results presented above include power transformer
losses modelled with a “European Business As Usual 2035” electricity prospective
mix. The sensitivity analyses carried out allow to study the impact of modelling
losses with either the French electricity mix or a 100% Wind Mix. Because electrical
losses are the primary source of impacts, the choice of the electrical mix used to
model them is a very important value.
4.2. LCA results – sensitivity analysis

After the rst iteration, a total of 11 sensitivity analysis have been performed within
the presented study.
Below an overview is given with the main results from 4 of the most in uencing key
parameters.
A rst sensitivity analysis was carried out in order to study the impact of the
variation in the offshore wind-farm load factor on the environmental footprint of the
substation. Two variants are studied and compared to the reference case (load factor
= 47.6%):

A low variant (load factor equal to approx. 43%)
A high variant (load factor equal to approx. 52%)
Figure 7 - Variations for the load factor (sensitivity analysis
The graph above (Fig.7) shows the variations in the impact of electric losses, which
are the only parameter impacted by the variation of the load factor. For the varying
parameter, and for the two analysed scenarios, the differences in results observed
are, for all impact indicators, of the order of 10% compared to in the reference
scenario. Losses remain a very signi cant element on the overall environmental
impact of the offshore substation, even with a low load factor.
In this study, RTE has chosen as the electricity production mix to be taken into
account to assess the environmental impact of electrical losses a mix of average
European production by 2035. This choice was made because the electricity network
France is interconnected to the European electricity grid, and that the electricity
mixes are changing in particular because of the energy transition. However, this
choice is not obvious.
The objective of this sensitivity analysis is to study the impact of the choice of the
electricity mix modelled for the losses on the results of the study. As indicated in the
diagram at Fig.8, the reference case in blue (forecast European Mix BAU 2035) was
compared with two variants:
A 100% wind-powered mix variant (in red)

The current French mix variant (in grey)

Figure 8 - Variations for Electric Mix (sensitivity analysis)
For the variant parameter, and for the two analysed scenarios, the differences in
results observed are quite signi cant. Thus, the choice of energy mix for loss
compensation has a great in uence on the results of the study.
This analysis therefore underlines the importance of clarifying the methodological

debate on how to account for and model them. Moreover, beyond this methodological
debate, it can be said that the main eco-design track for reducing the impacts linked
to electrical losses lies in outside the analysed perimeter of the OSS system.
The results of the LCA study are calculated from data related to an OSS equipped
with 2 transformers of a certain power capacity. In order to study the in uence of the
power to be transformed on the LCA and environmental impacts, two variants were
modelled, changing the main power transformers number (reference 2 transformers)
ref. diagram below (Fig.9):
An offshore substation variant with 1 transformer (in red);

An offshore substation variant with 3 transformers (in grey);
Figure 9 - Variations for number of main power transformers (sensitivity analysis)

Of course, all of the platform's data are impacted by these variants (structure, electric
equipment, transports, and losses) and changes taken into account. Results show
that the impacts of a substation equipped with a single transformer, weighted to the
MW installed, are 7% to 37% higher than for a substation equipped with two
transformers.
This is because although the structure and other equipment may be less important
to support a single transformer, these quantities are not divided by two. Similarly,
the differences in impact (weighted to the MW installed) between a substation
equipped with three transformers and the reference case with one transformer are
from -1% to -12% depending on the indicators. Thus, the variation in the number of
transformers or in power to be transformed does not signi cantly impact the results
of the life cycle analysis presented above.
The OSS offshore installation site and in particular its water depth also play a role.
Therefore, a sensitivity analysis case compares the 18 m water-depth reference case
(in blue) to 100 m water depth (in red) ref. diagram in Fig.10.
The different impacts brought by this variant mainly affect the OSS manufacturing
and installation phases, as well as its end of life. The graph below shows the
variations in impacts related to the manufacture, transport and end of life of the
substation, which are the parameters impacted by the 100 m water depth case.
Figure 10 - Variations for water depth at OSS (sensitivity analysis
The change in the OSS structure resulting from 100 m water depth mainly impacts
the manufacturing phase, in fact a greater quantity of materials is used to build the
structure. On all life cycle, the impacts of the 100m scenario are between 4% and 9%
greater than the impacts of the reference scenario depending on the indicators.
This preliminarily indicates that the water depth of the OSS installation site is an
important element of the environmental assessment, although additional work is
required to re ne these results.
4.3. Eco-design alternatives

The results of the investigation for potential alternatives led to identifying
suggestions for improving the OSS environmental footprint. Some of the possible
hints for improvement are:
OSS fabrication:
Minimize substructure and foundations (for example, by choosing a site as
shallow as possible, or with oating solutions); enhance weight optimisation /
reduction (mainly for steel and scour protection material, but also in general);
enable equipment life extension; increase of recovered and re-used
quantities at the end of the asset life.
OSS transportation:
Optimize global transportation routes / modes of transport; enable cleaner /
less polluting fuels for maritime and onshore transportation.
OSS operation:
Promote technologies with low electrical loss rates; improve Electricity Mix
and purchase greener electricity to compensate for losses; replace SF6 by
alternative gases and / or 0% leaks.
End of life:
Develop designs in order to increase reuse and recycling ratios
Promote effective methodologies and treatment to maximise actual recycling.
5. Conclusions
Within the context of offshore wind-farms grid connection projects, the LCA study
presented is a rst-in-its-kind to date, for several reasons: its comprehensive
perimeter of a full offshore substation; its iteration to nd alternative choices; its
systemic approach to design.
The study enabled RTE to gain an initial view of the environmental impacts of the
offshore substation, encompassing the construction phases (manufacturing,
transportation, installation) and operations phases (service, maintenance, removal,
recycling/disposal) for all its infrastructure and equipment.
The study allowed identifying the elements that contribute the most to these
impacts, and documenting a subject which remains little treated in the current
scienti c literature.
Iterations have been done in the study, in order to pick several alternatives in design
both for the electric power equipment and the infrastructure, depending on the
technical feasibility and the sensitivity on the overall environmental footprint.
In the detailed engineering phase, RTE will be able to re ne this study, by using
more precise data and comprehensive information, and remove the limits which
have been identi ed.
To go further, the ambition is to drive the company choices in design, supply chain,
construction, operations and maintenance aspects. On the path forward, the software
model will allow to manipulate the components based on the technological choices
made, and ascertain quantitatively the resulting effects.
Therefore, the future step will be to use the platform model such as an
environmental calculator during the detail engineering phase and for other next
projects. This environmental calculator will provide the elements allowing to decide

on the design choices in the same way as cost, time and risk analyses
are conventionally carried out. Intrinsically, the environment will be a part of the
equation in the same way as capital and operating expenditure, time and risk are
conventionally today.
Moreover, this method shall be extended to all the power transmission network. It
will be interesting to see the bene ts of a systematic LCA approach on such a full
scale system and study how global optimisation can be achieved through a wide
range of parameters.
References
1. PPE - Décret n° 2020-456 du 21 avril 2020 relatif à la programmation
pluriannuelle de l'énergie
2. EcoAct and Elys Conseil – OSS LCA Bibliographic analysis and methodology
report - 2020
Contact Author: D. BOTRUGNO 

Contact Author: M.NUNES 

CSE 024
Case study of dc-MMC interconnecting two

HVDC lines with different grid topologies
AUTHORS
D. GOMEZ A. - SuperGrid Institute & Universitat Politècnica de Catalunya

K. SHINODA, J.D. PAEZ, F. MOREL, P. DWORAKOWSKI - SuperGrid Institute
M. CHEAH-MANE, O. GOMIS-BELLMUNT - Universitat Politècnica de Catalunya
Summary
Dc-dc converters for high voltage direct current (HVDC) applications are interesting
devices for the development of multi-terminal and dc grid schemes. They can adapt
the differences between the dc systems (voltage level, technology and/or line
topology) and provide additional protection, controllability and redundancy services.
This paper presents a case study of a non-isolated dc-dc converter interconnecting
the NordLink and Cobra cable projects. This interconnection requires a dc-dc
converter to adapt the dc voltage and the line topology (the rst project being a ±525
kV rigid bipole while the second is a ±320 kV symmetric monopole). The case study
has been modelled in Matlab/Simulink and several scenarios are simulated. A rst
set of 8 simulations change the power ow directions verifying the adequate
behavior of the dc-dc converter and HVDC links. Based on the previous simulations,
additional 16 cases are simulated to verify the system behavior during faults. Two
faults are proposed: a pole-to-ground on the positive pole of the rigid bipole line and
another on the positive pole of the monopole. It is shown that the dc-dc converter can
interconnect two existing lines without changing their initial control strategy. For the
normal operation simulations, a fast power variation through the dc-dc converter is
tested proving that the line protections are not triggered. The power perturbations
created by the dc-dc converter are small enough to keep the system under normal
operation conditions. During the fault simulations, the dc-dc converter is able to
isolate the healthy side from the fault disturbance ( rewall capability). The healthy
dc side observes a power disturbance (loss of power ow through the dc-dc
converter) without triggering the protections, allowing the continuous power
transmission and regaining steady state conditions. The dc-dc converter employs full
bridge submodules to be able to block the fault currents, without the presence of dc
circuit breakers.

KEYWORDS
dc-dc HVDC converters - dc-MMC - grid topologies - multi-terminal dc grid - modular multilevel
converters
1. Introduction
High voltage direct current (HVDC) transmission lines are currently used to transmit
high power over long distances. The interest in interconnecting two or more HVDC
lines has risen in the latest years inspiring multiple publications and a CIGRE
working group (B4.76) dedicated to the study of dc-dc converters for HVDC
applications.
HVDC transmission system does not have an established standard yet, hence, the
installation features vary from one project to another, such as the operation voltage,
the technology (line commutated converter or voltage source converter), the line
topology (e.g. bipole, monopole), the grounding strategy (e.g. start point reactor), the
protection scheme or the conductor type used (cable or overhead lines). Dc-dc
converters are important devices for the development of future dc grids [1] as they
provide the interface needed to interconnect HVDC systems with different
characteristics. Several HVDC dc-dc converters have been proposed in the literature
to interconnect two dc systems [2], but the literature focuses mainly on the
interconnections between two dc systems with different voltage levels [3], while the
impact of their grid topology is not often considered.
There are four different grid topologies for HVDC transmission systems: asymmetric
monopole (Fig. 1a), symmetric monopole (Fig. 1b), bipole (Fig. 1c) and rigid bipole
(Fig. 1d) [4]. Each topology behaves differently in case of a fault. There are different
possible faults: faults pole-to-ground, pole-to-pole, or internal faults in a converter.
Both monopoles and the rigid bipole must stop the power transmission during any
fault and they remain inoperative until the fault is cleared. Compared to the
monopoles, the rigid bipole can be recon gured to isolate a fault on a converter. Then
it can use the healthy conductors and converters to exchange, up to, half of the rated
power. However, the rigid bipole loses the complete power exchange capacity in case
of a fault in a conductor. In case of a pole-to-ground fault, the symmetric monopole
and the rigid bipole have a pole displacement leading to an increased voltage (pole-
to-ground) on the healthy pole. The symmetric monopole can experience voltages up
to 2 p.u., whereas, voltages around 1.4 p.u. have been reported for a given rigid
bipole application [5]. The main advantage of the bipole is its inherent redundancy.
Contrary to the rigid bipole, the bipole continues the power transmission, without
interruption, for any fault except the pole-to-pole fault. The bipole use the healthy
pole and the metallic/ground return to exchange up to half the rated power without
being subject to voltage or currents disturbances.

Figure 1 - HVDC grid topologies. (a) asymmetric monopole (b) symmetric monopole (SyM) (c)bipole
with metallic or ground return and (d) rigid bipole (RB)
To interconnect two dc systems with different grid topologies, a dc-dc converter is

needed. The dc-dc converter must respect the operation limits of each system
throughout any operation condition, including faults. These operation limits can vary
depending on the installed equipment, the protection strategy and the fault
detection algorithm. This paper focuses on the operation limits given by a basic fault
detection algorithm. The interconnection of two dc systems with different grid
topologies (rigid bipole and symmetric monopole) is considered in this paper. To
interconnect the lines, a non-isolated dc-dc converter is considered, here called
exible dc-MMC. The converter is based on the one presented in [6], [7]. It offers
lower losses and cost compared to the isolated dc-dc converter front-to-front [8], [9].
The study evaluates the dc-dc converter behavior and its impact on the HVDC
systems in normal operation and during faults.
The paper starts by presenting the main characteristics of the HVDC lines to
interconnect. Section 3 presents the topology and control of power converters (ac-dc
and dc-dc). Section 4 presents the parameters of the simulation model. The
simulation results are shown in section 5 and the conclusions are given in section 6.
2. Proposed case study

This paper investigates the use of a dc-dc converter interconnecting two lines with
different grid topologies based on a possible case proposed in [4]: the
interconnection between the NordLink and Cobra Cable projects. The NordLink
project has 54 km of underground cable and 516 km of submarine cable connecting
Norway and Germany. This HVDC link is based on a rigid bipole (RB) rated at 1400
MW with ±525 kV [10]. The Cobra Cable project has 325 km of submarine cable
interconnecting Denmark and the Netherlands. This HVDC link is based on a
symmetric monopole (SyM) rated at 700 MW with ±320 kV [11]. The approximated
geographical layout is presented in Fig. 2. The interconnection demands a dc-dc
converter able to adjust the dc voltage and the grid topology between the two
projects.

The possible bene t of this interconnection is the increased power ow exibility
between all the interconnected countries: Netherlands, Germany, Denmark and
Norway. The installation of a dc-dc converter enables the direct power exchange
between Norway and Netherlands, not possible in the initial system. The dc-dc
converter can also increase the redundancy of the interconnection. After a line or
converter fault, the interconnected system can recon gure itself to isolate the
damaged segment allowing the power exchange between the healthy sections. For
example, in case of a fault between the dc/dc converter and Denmark (or in the ac/dc
converter in Denmark), a power exchange with Netherlands-Norway-Germany could
be possible after the fault isolation. In the initial system, the same fault will isolate
the Netherlands from the Norway.
Figure 2 - Map of the North Sea region presenting NordLink and Cobra cable. The dc-dc converter is
proposed at the crossing point
3. Converter topologies and control

The proposed case study considers two types of HVDC converters: ac-dc and dc-dc
converters. The considered ac-dc converters are modular multilevel converters
(MMC). The dc-dc converter is a variation of the dc-MMC presented in [6]. Some
details are presented in this section.
3.1. Ac-dc modular multilevel converter

The MMC is shown in Fig. 3, it has two arms per phase (upper and lower). Each arm is
composed of a series connection of half bridge submodules (HBSMs) and an inductor.
At the connection point between the upper and lower arm, there is an ac output port.
The converters used here have 3 legs for a three-phase connection to the ac grid.
The MMC controller is designed in line with [12]. A general control diagram is
presented in Fig. 4a. Three energy controllers (total, horizontal and vertical) manage
the energy by setting the current references (Iref diff and Iref ac), which are a linear
combination of the upper and lower arm currents Iup and Ilow. The control changes
the ac current reference depending on the control mode, either voltage control or

power control. The voltage controllers were tuned to have a second order response
with a time constant of 100 ms and an overshoot of 10%. The converter control
adopts a virtual capacitor controller (VCC) as presented in [12], which is a
proportional control that helps
to support the dc voltage variations using the energy in the converters: energy
stored in the SMs capacitors. The energy in the converter varies proportionally to the
dc voltage uctuation and the virtual capacitor coef cient value, here set to 1. This
coef cient is used to calculate the voltage and energy controllers.
Figure 3 - Circuit diagram of the ac-dc modular multilevel converter: submodule (SM), arm inductor (L)
and half bridge submodules (HBSM)
The power control in the rigid bipole is modi ed (Fig. 4b): both MMCs (positive and
negative pole) control the total energy with the ac current and have a common dc
terminal voltage ΔVdc to control the dc power [13]. The station controlling the
voltage implements the control presented in Fig. 4a including VCC.
Figure 4 - MMC control diagrams (a) The general control strategy for power or voltage control mode

Figure 4 - MMC control diagrams (b) The terminal control for the rigid bipole (ungrounded station)
3.2. Flexible dc-dc modular multilevel converter

The dc-dc converter used for the proposed case study is a modular structure
presented in Fig. 5. Each phase of the converter is composed of 3 arms (upper,
middle and lower). These arms contain a series connection of submodules (SMs)
either half bridge (HBSM) or full bridge (FBSM) and an arm inductor (L). The FBSMs
are used to have negative voltage outputs and to block the dc fault currents. The dc-
dc converter needs enough FBSMs to block the fault interactions between the two
interconnected dc systems. In this study, the upper and lower arms use 400 FBSMs
each. The converter has an output passive lter (Lout) allowing the ac power to
circulate between phases and other arms. These interactions are needed to keep the
internal energy balanced [7], [14]. Although Lout works as an ac lter, the control
strategy keeps the phase currents balanced to avoid ac components on the dc side.
The high voltage side (left side in Fig. 5) is connected to the rigid bipole while the low
voltage side (right side in Fig. 5) is connected to the symmetric monopole line. The
number of phases can change depending on the converter rated power and the
semiconductors current rating. A dc-dc converter with three phases is considered in
this paper.
The converter is controlled to keep the internal energy constant, while a power is
transferred between dc systems. The control strategy is a cascaded structure with
the high level control following the energy references and the low level controlling
the currents. There are three energy controllers per converter phase: total energy
(Σ), total difference (ΔT) and the difference between the upper and lower arm (Δu-l).
The internal currents are controlled to follow the references given by the energy
controllers, the pole imbalance (on the monopole side) and the power reference, as
shown in Fig. 6. In this work, a power control mode is implemented, but additional
controllers can be added to control the voltage on the dc sides.

Figure 5 - Flexible dc-MMC for interconnections between different grid topologies
Figure 6 - Control strategy used for the dc-dc converter
4. Simulation model
Figure 7 shows the schematic of the simulation model implemented in
Matlab/Simulink. The 7 converters were implemented using the parameters
presented in Table 1. The converters use average arm models, including the blocked
state, depending if they are composed by HBSM [15] or FBSM [16]. The cables are
modelled using the wideband model from the Best Paths project [17]. The model
considers two reference points: the rst on the dc side of the RB, which is grounded
through a resistance (RG) of 0.7 Ω as presented in [5]; the second, on the ac side of
the SyM line through a star point reactor as shown in Fig. 7 [18, p. 291].

Figure 7- Proposed case study interconnecting a rigid bipole (upper side) with a symmetric monopole
(lower side). The currents present the positive power flow direction in the lines and through the dc-dc
converter. The squares adjacent to each ac-dc converter represent the ac circuit breakers
Table 1 - Converters parameters
The ve converter stations implement a fault detection algorithm, which uses local
voltage and current measurements to detect a fault. The fault detection algorithm
blocks the converter before the arm current reaches 3 kA, as the complete case study
was designed using IGBT valves rated at 1.5 kA and 3 kV. After blocking the
converter, it is assumed that the fault current circulates through the by-pass
thyristors integrated within the SMs [18]. In addition, the algorithm trips the station
ac circuit breakers (ACCB) to stop the fault current contribution from the ac side. The
fault detection algorithm has a 300μs delay [19] to emulate the detection,
communication and tripping time inside a converter station. An additional 100ms
delay is added to the ACCB opening signal. With regard to the voltage
measurements, the algorithm triggers the block signal if the measurements deviate

20% from their nominal values. The voltage thresholds in the dc-dc converter are set
to 15% of deviation, to block the converter before the tripping of the protections in
the lines. This last feature allows the dc-dc converter to act as a rewall i.e.
preventing the spreading of fault disturbances from one side to the other.
5. Simulation results
The proposed case study interconnecting two HVDC lines using a dc-dc converter
may have multiple power ow distributions around the system. Figure 7 shows the
positive convention for the dc currents, which are also used for the power direction.
The positive power direction in the lines are from the dc side towards the power
controlled stations, whereas the positive convention in the dc-dc converter is from
the RB to the SyM line. Eight power ow variations are proposed to test the system
during normal operation, i.e. without faults. Based on these eight power ow
scenarios, the fault behavior is analyzed considering 2 different fault locations per
case, which gives a total of 16 fault simulations. The considered fault locations are: a
fault on the positive pole of the rigid bipole (P-Ctrl RB+) and a fault on the positive
pole of the V-Ctrl SyM station, as shown in Fig. 7. The details of the power ow
directions and fault locations are organized in Table 2.
Table 2 - Variations of the case study, where the positive conventions (+) are shown in Fig. 7. In the RB
and SyM lines from right to left (towards the P-Ctrl stations) and in the dc-dc converter from the RB to
the SyM
The simulation starts with nominal voltage and power reference on the RB and SyM
lines, 1000 MW and 300 MW respectively. After reaching the steady state, a power
reference is sent to the dc-dc converter to exchange power (400 MW) between dc
systems in both directions. The simulation results in normal operation are shown in
Fig. 8, for case 1: positive power ow in the lines and bidirectional exchange through
the dc-dc converter. The results obtained for the other 7 cases are similar: the
uctuations presented in Fig. 8 are representative for all tested normal operation
cases. The power exchange measured at the ve converter stations are presented in
Fig. 8a. It can be seen that the stations in voltage control mode change their power
exchange to compensate the power through the dc-dc converter (magenta line), while
the stations in power control keep their power exchange invariant as their references
have not been changed (solid lines).
Figure 8b shows the voltage variations pole-to-ground on each converter. These

perturbations depend on the voltage control tuning on each line (time response of
100 ms) and the rate of change in dc-dc converter power reference (2.3 GW/s, i.e. 3.3
p.u./s). In this case study, the voltage perturbations do not exceed 6% of the nominal
line voltage, but they can be reduced if a decreased slope in the dc-dc converter
power references is used.

Figure 8c shows the variations in the average total energy stored in every converter.
In this case, the power controlled station in the SyM line has the higher energy
variation (11.4%) followed by the voltage control station in the same line. These
energy variations are the consequence of the VCC, which helps to keep the dc voltage
regulation using the converter internal energy. For the converters in the RB line, the
VCC is implemented only in the V-Ctrl station. The P-Ctrl station controls the total
converter energy with the ac power (Fig.4b) therefore, P-Ctrl in RB keeps its energy
constant. Similar to the voltage variation, these energy disturbances can be
attenuated by reducing the slope in the dc-dc converter power reference, but for this
paper the slope remains unchanged to test the systems responses to fast power
transients.
Figure 8 - Simulation results in normal operation for the case 1. (a) Power per station. (b) Pole-to-
ground voltage, positive (+) or negative (-) measured at the different converter terminals in the system
(the negative poles were inversed to simplify the visualization). (c) Total energy in the converters
Regarding the fault simulations, the starting sequence is the same, but the dc-dc
converter only exchanges power in one direction: positive from RB to SyM and
negative in the opposite direction. After reaching the steady state, a fault of 1 Ω to
ground is simulated (at 2s), at the terminals of the P-Ctrl RB+ or V-Ctrl SyM as shown
in Fig. 7. The results presented in Fig. 9 are the worst cases per station found when a
fault in the RB line is simulated (odd cases of Table 2). During the fault, the voltage in
the faulted pole drops to zero as shown in Fig. 9a. At the fault instant (t=2s), the
negative pole in the RB has an increased voltage reaching -880 kV for the 15th case
(see Fig. 9b). The steady state voltage for this pole stabilizes at -762 kV, which
represents 1.45 p.u., similar to the results found in [5]. On the SyM side (Fig. 9b-c),
the voltages do not exceed ±10% of the nominal voltage for any case, which agrees
with the initial control design.

Figure 9 - Dc voltages pole-to-ground during and after the fault on the RB side (worst cases). (a)
positive pole in RB. (b) Negative pole RB. (c) Positive pole SyM. (d) Negative pole SyM
Figure 10 presents the higher dc currents measured at the RB (Fig. 10a), the SyM
(Fig. 10b) and the dc-dc converter (Fig. 10c) terminals for the cases when the fault is
on the RB side. Figure 10a shows a peak of 8 kA for the currents measured in V-Ctrl
RB+ converter. It is worth mentioning that the converter protection strategy blocks
the MMC under 400μs after the fault. The currents observed in V-Ctrl RB+ continue
to rise until the ACCB is opened: after 2.1 s, which illustrates how the ac currents
feed the fault in that pole. On the SyM side (Fig. 10b) the currents have a
perturbation due to the loss of power from/to the dc-dc converter (t = 2s). After the
transient, the currents reach the operation point where the voltage controlled
stations exchange the same power as the power controlled stations (behavior
expected without the use of dc-dc). Figures 9b-c and 10b illustrate that the dc-dc
converter can act as a rewall during a fault in a RB, as the healthy link continues
normal operation once the dc-dc converter is blocked, irrespectively of the state of
the other link.
Figure 10 - Dc currents during and after the fault on the RB side (worst cases). (a) currents in all four
MMCs in the RB line. (b) Currents measured in the SyM. (c) Currents measured in the dc-dc

Figure 11 shows the worst voltages measured at the stations, for the simulations
when the fault is on the SyM line (even cases in Table 2). The voltages on the RB
exceed the 10% of variation with respect to the nominal value for the cases 8 (Fig.
11b), 10 and 12 (Fig. 11a). These uctuations are consequence of the RB control
scheme, which is sensitive to the variations in the dc side due to the terminal control
in the P-Ctrl station. The voltage uctuations depend on the rate of change of the
power delivered by the dc-dc converter. These variations can be reduced if the dc-dc
converter reduce the disturbances during the fault. Reference [14] present a control
strategy that allows the converter to control the fault current, decreasing the
magnitude of the disturbances. On the SyM side, the positive pole voltage drops to
zero (Fig. 11c), while the negative pole presents an overvoltage of 2 p.u. (Fig. 11d).
The results were obtained without surge arresters, but the implementation of
overvoltage protections can reduce the voltage peak and steady state values.
Figure 11 - Dc voltages pole-to-ground during and after the fault on the RB side (worst cases). (a)
positive pole in RB. (b) Negative pole RB. (c) Positive pole SyM. (d) Negative pole SyM
Figure 12a shows the currents in the RB and how they recover steady state values
after the fault. Similar to the results presented in Fig. 10, the power and voltage
controlled stations nd the operation point where the dc-dc is absent. Figure 12b
presents the dc currents in the SyM and how they are blocked in a short time (less
than 50 ms). The currents through the dc-dc converter are presented in Fig. 12c. The
dc-dc converter is able to isolate the RB line during a fault in the SyM allowing the
rst to regain steady state operation without triggering any protection.

Figure 12 - Dc currents during and after the fault on the RB side (all even cases on Table 2). (a) currents
in all four MMCs in the RB line. (b) Currents measured in the SyM. (c) Currents measured in the dc-dc
Figure 13 presents the energy in the dc-dc converter during and after the 16 faults
cases of Table 2. Figure 13a presents the cases for a fault on the RB side. In these
scenarios the energy variations are near 1% of the nominal value (30.24 MJ). For the
cases where the fault is located on the SyM (Fig. 13b), the energy in the converter
increases up to 110% as a consequence of the FBSMs charging as they stop the fault
current.
Figure 13 - Energy in the dc-dc converter for (a) a fault in the RB (odd cases of Table 2) and (b) a fault in
the SyM (even cases of Table 2)
6. Conclusions
Dc-dc converters are needed for the interconnection between dc systems with
different characteristics. The present publication explores the hypothetical
interconnection between the NordLink (rigid bipole - RB) and Cobra Cable
(symmetric monopole – SyM), using a non-isolated dc-dc converter.
The case study was modelled and simulated in Matlab/Simulink. The power ow in
the lines and through the dc-dc converter was changed to simulate 8 possible normal
conditions. These initial simulations validated that the exchange of power between
the two lines does not trip any protection. After validating the converter behavior in
normal operation, a set of 16 fault cases were simulated. Satisfactory results were
obtained for the dc-dc converter rewall capability: in case of a fault in a line, the dc-
dc converter is able to stop the fault propagation to the healthy system. This rewall
capability was validated independently of the power ow direction or the fault
location. If enough FBSMs are installed in the dc-dc converter, the rewall capability
can be ensured without the presence of dc circuit breakers.

The interconnected lines do not require major adjustments to operate in an
interconnected system using a dc-dc converter. The control strategy implemented in
the dc-dc converter can be adapted to guarantee the normal operation of the initial
lines, and its protection strategy can be design to operate before the line protections.
A fast power exchange between dc lines was tested in this paper disturbing the
operation of the lines without triggering the fault protections. A slower rate of change
can be implemented to reduce the disturbances on the system.
The case study proposed in this paper did not change the grounding strategy and
reference points of the initial links. A dc reference is found on the RB and an ac
reference on the SyM line. In this particular case, no problems were found, but
further studies must be carried to check the impact of multiple reference points on
the protection strategy and isolation coordination.
Further studies should evaluate the impact of the ac faults on the dc-dc converter
control strategy and fault detection algorithm. Additional fault studies should be
done to assess the dc-dc converter behavior under different fault conditions.
Different types of faults and locations should be tested. Supplementary studies can
be done for the degraded operation, where a fault in the interconnected system leads
to the isolation of a portion of a line (e.g. a three-terminal system or the RB
recon guration into an asymmetric monopole - isolating the faulted converter).
Acknowledgment
This work has been supported by a grant overseen by the French National Research
Agency (ANR) as part of the “Investissements d’Avenir” Program (ANE-ITE-002-01).
References
1. CIGRE WG. B4.52, “HVDC Grid Feasibility Study.” CIGRE, Apr-2013 [Online].
Available: https://ecigre. org/publication/533-hvdc-grid-feasibility-study
2. J. D. Páez, D. Frey, J. Maneiro, S. Bacha, and P. Dworakowski, “Overview of DC–DC
Converters Dedicated to HVdc Grids,” IEEE Transactions on Power Delivery, vol.
34, no. 1, pp. 119–128, Feb. 2019.
3. D. Jovcic, P. Dworakowski, G. Kish, A. Jamshidi Far, A. Nami Abb, A. Darbandi, and
X. Guillaud, “Case Study of Non-Isolated MMC DC-DC Converter in HVDC Grids,” in
CIGRE Symposium 2019, 2019.
4. D. Gómez, J. D. Páez, M. Cheah-Mane, J. Maneiro, P. Dworakowski, O. Gomis-
Bellmunt, and F. Morel, “Requirements for interconnection of HVDC links with DC-
DC converters,” in IECON 2019-45th Annual Conference of the IEEE Industrial
Electronics Society, 2019, vol. 1, pp. 4854–4860.
5. M. Goertz, C. Hirsching, S. Wenig, K. M. Schäfer, S. Beckle, J. Reisbeck, M. Kahl, M.
Suriyah, and T. Leibfried, “Analysis of overvoltage levels in the rigid bipolar MMC-
HVDC con guration,” in 15th IET International Conference on AC and DC Power
Transmission (ACDC 2019), 2019, pp. 1–6.
6. G. J. Kish and P. W. Lehn, “Modeling techniques for dynamic and steady-state
analysis of modular multilevel DC/DC converters,” IEEE Transactions on Power
Delivery, vol. 31, no. 6, pp. 2502–2510, 2015.
7. G. J. Kish, M. Ranjram, and P. W. Lehn, “A Modular Multilevel DC/DC Converter
With Fault Blocking Capability for HVDC Interconnects,” IEEE Transactions on
Power Electronics, vol. 30, no. 1, pp. 148– 162, Jan. 2015.
8. T. Lüth, M. M. Merlin, T. C. Green, F. Hassan, and C. D. Barker, “High-frequency
operation of a DC/AC/DC system for HVDC applications,” IEEE Transactions on

Power Electronics, vol. 29, no. 8, pp. 4107–4115, 2013.
9. F. Sasongko, M. Hagiwara, and H. Akagi, “A front-to-front (FTF) system consisting
of two modular multilevel cascade converters based on double-star chopper-
cells,” in 2013 1st International Future Energy Electronics Conference (IFEEC),
2013, pp. 488–493.
10. hitachi ABB Power Grid, “NordLink.” [Online]. Available:
https://www.hitachiabbpowergrids. com/references/hvdc/nordlink
11. B. Tourgoutian and A. Alefragkis, “Design considerations for the COBRAcable
HVDC interconnector,” in IET International Conference on Resilience of
Transmission and Distribution Networks (RTDN 2017), 2017, pp. 1–7.
12. [12] K. Shinoda, A. Benchaib, J. Dai, and X. Guillaud, “Virtual capacitor control:
mitigation of DC voltage uctuations in MMC-based HVdc systems,” IEEE
Transactions on Power Delivery, vol. 33, no. 1, pp. 455– 465, 2017.
13. C. Hirsching, M. Goertz, S. Wenig, S. Beckler, M. Suriyah, and T. Leibfried, “On
control and balancing of MMC-HVdc links in rigid bipolar con guration,” in 15th
IET International Conference on AC and DC Power Transmission (ACDC 2019),
2019, pp. 1–6.
14. M. Cheah-Mane, J. Arevalo-Soler, E. P. Araujo, and O. Gomis-Bellmunt, “Energy-
based control of a DC Modular Multilevel Converter for HVDC grids,” IEEE
Transactions on Power Delivery, 2019.
15. H. Zhang, D. Jovcic, W. Lin, and A. J. Far, “Average value MMC model with accurate
blocked state and cell charging/discharging dynamics,” in 2016 4th International
Symposium on Environmental Friendly Energies and Applications (EFEA), 2016,
pp. 1–6.
16. F. Xinkai, Z. Baohui, and W. Yanting, “Fast electromagnetic transient simulation
models of full-bridge modular multilevel converter,” in 2016 IEEE PES Asia-
Paci c Power and Energy Engineering Conference (APPEEC), 2016, pp. 998–1002.
17. B. Paths, “Best Paths, Transmission for sustainability.” Best Paths [Online].
Available: http://www.bestpaths-project.eu/
18. K. Sharifabadi, L. Harnefors, H.-P. Nee, S. Norrga, and R. Teodorescu, Design,
control, and application of modular multilevel converters for HVDC transmission
systems. John Wiley & Sons, 2016.
19. S. Dennetière, S. Nguefeu, H. Saad, and J. Mahseredjian, “Modeling of modular
multilevel converters for the France-Spain link,” Star, vol. 2, no. 3, p. 4, 2013.
Contact Author: D. GOMEZ A. 

CSE 024
Information Synergy: Integrating data

across disciplines and functions
AUTHORS
A.C. WEST - SUBNET Solutions Pty Ltd
Summary
The integration of new technologies is an on-going element of Grid Modernization.
New technologies receive intense focus, sometimes at the expense of emphasis or
awareness in other areas. Grid automation is an inherently complex task, bringing
together many specializations. Utilities have evolved internal organizational
structures that recognize and support the multi-disciplinary approach needed for
successful automation implementation. IEC 61850 and disruptive Internet of Things
(IoT) technologies present new paradigms that can challenge these traditional
structures and adoption of such technologies may best be supported by refactoring
traditional work ows. In such restructuring, emphasis is often placed on addressing
new requirements and integrating new methods of engineering systems. In doing so,
it is easy to overlook on-going needs of existing systems and functions or to neglect
to take the opportunity to review whole-of-life information management of the
system. For some systems, additional design effort can yield dividends in additional
functionality or reduced operational expense.
The introduction of IEC 61850 brings a need for closer cooperation between
protective relay engineering, substation automation, SCADA, eld test and
communication system functions. For some utilities this can represent a signi cant
restructuring of work ows and responsibilities, thus presenting an opportunity for
improved cooperation and information sharing between historically-independent
functional groups.
Many functions interact with grid automation systems: Planning,

architecture/design, con guration, commissioning, testing, operation, protection,
maintenance, asset management, etc., with overarching requirements such as
cybersecurity and corporate policy objectives.
When introducing new technologies, it is worthwhile considering what additional

integration and data sharing may be possible. In many cases, bene ts may be
obtained by providing data access for a larger number of functional groups within

the utility. In performing this analysis, the policies and implementation of
cybersecurity of the system and the control of access to data needs to be considered.
Each user with a legitimate business need to access data should be supported to
access that information without undue impediment, while simultaneously restricting
their access or ability to modify information not related to their job functions.
An integrated, holistic approach to data access for operational and corporate

purposes that integrates strong cybersecurity and role-based access for each user
can enable operational ef ciencies and lead to greater system resilience.
Systems are becoming more complex and offer a growing number of capabilities. It is
necessary to manage those capabilities in a manner that simpli es each person’s
role and assists them to avoid mistakes that negatively impact system performance.
An integrated method of managing equipment and securing access to information is
one tool that can aid this process.
This paper serves to remind the reader to implement a multi-disciplinary approach

to consider all information uses associated with a system, not only to manage the
design, test and commissioning process; but also to facilitate on-going system
operation and support of information for non-operational functions.
KEYWORDS
Cybersecurity - Grid automation - Device management
1. Introduction
The evolution of the grid, the development and inclusion of new control devices and
the foreseeable integration of IoT devices each expand the amount and diversity of
information used in grid management.
Recent introductions such as IEC 61850 and Phasor Measurement Units are
examples of new technologies which provide new operational data, but also depend
on speci c functionalities of the associated communications and con guration
management systems.
New sources of data can enable new applications such as system stability analysis
from phasor measurements or allow new business practices such as performing
maintenance based on asset monitoring instead of on a periodic basis.
2. Business drivers
Many utilities have organizational structures that have separate responsible entities
for functions such as power system planning, control system design, protection,
communication, system engineering (con guration, testing and maintenance),
operations, asset management, disturbance analysis, etc. These separate areas of
responsibility and their operating procedures have evolved with the evolution of the
grid. IEC 61850 integrates aspects of several of these areas of responsibility. Some
utilities have found that adoption of these new technologies can stimulate or even

mandate organizational restructuring in order to coordinate the engineering and
operation of the new systems. Sometimes new multi-disciplinary teams need to be
formed, bringing expertise from separate areas together.
The adoption of new technologies and standards typically provides value in terms of
offering additional functionality coupled with manageable additional engineering to
achieve that functionality. The adoption of new technologies does not replace the
need to properly engineer system design, however, they may provide tools to
simplify that task or parts of it. The framework provided by complex standards
provide for consistent and interoperable interpretations of data, commands and
con guration information. Without such common frameworks, the implementation of
complex functions would require additional engineering and expense.
Electric utility automation systems are typically an integration of multiple

subsystems (e.g. substation automation, SCADA, asset management, distribution
automation, etc.). Traditionally, each of these systems have been managed by a
separate responsible entity. While they each may have been operated more-or-less
independently, there are synergies between them that make information sharing
between systems bene cial. In many cases, speci c integrations are performed to
allow data sharing between subsystems to provide functional bene ts to the utility. It
is extremely unusual that all subsystems are updated at the same time, therefore
the interaction between those subsystems needs to be considered whenever
upgrading one of them.
3. Design considerations
When integrating new technology or new functionality, the impacts of that technology
or functionality on other aspects of the utility business need to be considered, so as
to ensure that all services and support needed by the new technology (e.g.
communication system updates, additional staff training, etc.) are also incorporated
as part of the project.
Further bene t may also be derived if it is feasible to also integrate the new
technology with other stakeholder use cases. For example: The provision of
enhanced communication services may permit access to additional data for other
purposes such as asset health monitoring, remote device access for engineering and
maintenance, etc. If the needs of all utility divisions are considered as part of any
project, the overall capability and ef ciency of the utility may be enhanced by using
the opportunity to also integrate support for the various data use cases of each group
of users. At a minimum, any new automation project should consider on-going
system operation and management needs as part of the design process. This may
consider aspects such as:
on-going self-monitoring and testing

access for maintenance functions
data retrieval for fault analysis
support for automatic or remotely-operated testing
cybersecurity objectives and/or compliance
4. Architectural considerations

The provision of data for a diversity of applications moves the utility from a
traditional model where communication access to the automation system is only
provided for operational purposes (usually to a SCADA Control Centre) to a model
where multiple users are provided access to a much wider range of data. In this
migration, the system becomes exposed to a wider range of people and the need to
manage and control that access also changes. The utility’s cybersecurity policies
may need review to ensure that they are applicable to the various kinds of access
and new policies and procedures may need to be implemented. In most cases, this
needs involvement of the utility IT group, the operational system design team, the
maintenance engineering team and other stakeholders. Each category of user may
need speci c access to different kinds of data that become available from the
automation system, and should be protected from inadvertently (or deliberately)
accessing or changing information that they do not require or do not have a need to
manage. In general, a system that supports role-based-access will typically be
needed to meet cybersecurity policy objectives.
The need to ensure isolation of the automation system equipment from the corporate
IT network (and, beyond that, the public internet), while also allowing controlled
access to data from various users will typically lead to a review of the communication
system architecture requirements. Guidance on this is available from various
sources, e.g.: NIST SP-800-82 [1]. When controlling access into an operational
automation device network, it is usually appropriate to use architectures that
conform to a “Zones and Conduits” model as described in IEC 62443 [2] where data
between a less trusted network (the “Enterprise Zone”) and a more trusted network
(the automation system zones) is only permitted through controlled paths (the
conduits) and is monitored and managed to ensure that only permitted traf c passes
between zones. Each utility should make its own determination regarding such
architectural analysis and how its cybersecurity policies can be implemented while
also providing the required functionality. In general, zones are rewalled from each
other and care should be taken to avoid simply opening “holes” through the rewall
for automation system traf c: More sophisticated architectures with demilitarized
zones and jump servers are typically required that do not permit direct access from
one zone to another.
When the various kinds of substation and distribution devices are included, together
with the various communication systems used to access those sites and the
integration with data collected for various operational and non-operational purposes
is directed to the users who have responsibility for that information, an architecture
might result such as that shown in Figure 1. Different kinds of data will typically use
different protocols: Operational data might use IEC 60870-5 or IEC 61850; fault
records might be collected as COMTRADE les; maintenance engineering will usually
use a proprietary vendor protocol. The integrated system is responsible for ensuring
that each user is provided access to the information required to perform their role
and to isolate that data from any impact caused by other users or functions.

Figure 1 - Typical Utility Communication Architecture
Cybersecurity regulations for electric utilities vary from jurisdiction to jurisdiction.

In North America, the NERC CIP standards dictate particular requirements for the
automation system and policies and procedures for staff. The European NIS Directive
identi es a different approach that a utility may be required to adopt and focuses
more on monitoring of the control system to enterprise interface for malicious
activity. There is merit in each approach and they may be considered to complement
each other. Implementing a utility’s own internal cybersecurity policies may require
compliance to local regulations as well as additional cybersecurity controls.
Cybersecurity can require system-level design and implementation considerations.
This itself can require a multi-disciplinary approach.
5. Standards
IEC 61850 is now seen to provide functional bene ts and is expanding its reach to
wider segments of the electric utility space, but it still does not address all functional
requirements.
IEC 61850 within Substation Automation Systems has demonstrated an ability to

support complex system architectures and a wide range of substation
automation functions. The concept of a single con guration de nition (using
System Con guration Language: SCL) for the whole substation has bene ts for
engineering, on-going management and documentation. SCL can be extended to
include “private” data covering non-IEC 61850 equipment in the substation. For
systems above a certain level of complexity, this can be more cost effective than
traditional methodologies.
IEC 61850 design ts well with the needs of substation automation, especially
the object modelling and introduction of additional protocols:
GOOSE messaging for fast, reliable coordination of protection signals between
relays
Sampled Values (SV) for the sharing of CT and PT data across communication
networks rather than by directly wiring these signals to each relay
IEC 61850 now de nes “Technical Reports” (akin to “Application Notes” or
guidelines for the implementation of speci c functionality) covering a wide range
of utility functions, but the list of these functions is not exhaustive.

IEC 61850 has changed how some functions are performed. Signals that might
previously appeared as discrete wired terminals on a relay (where they can be
independently monitored or isolated from other devices for test purposes by lifting
links) now appear as data on the network. This requires alteration of design, test and
maintenance practices and corresponding training in the new techniques.
Maintenance operations such as isolating operator controls need to be

considered as part of system design and may require speci c programming of
tools to automate these functions rather than just listing a set of steps on a
manual checklist. In many cases, maintenance procedures should be considered
as part of the system design process: The ability to monitor and manipulate
automation system signals for test and maintenance purposes will sometimes
depend on that access being speci cally identi ed as a functional requirement
during system design. This may be an iterative process as the additional
maintenance requirements are identi ed during system design.
New maintenance tools become possible to inject test signals (both GOOSE and
SV) to verify automation functions. IEC 61850 de nes methods to allow test
signals and operational signals to coexist on a network, permitting various forms
of in-service testing with minimal system disruption. Remote testing becomes
possible if appropriate test equipment is included as part of the deployed
installation.
A caveat: While speci ed in the standard, test functionality is not uniformly

implemented. During system design, the actual behaviour of devices in the
various data simulation and test modes must be veri ed to ensure that actual
operation is as expected. Failure to do this correctly can compromise personnel
safety!
Various functions remain out-of-scope for IEC 61850. For example: A protection relay
may have settings and settings groups that can be modi ed remotely, but not
protected by the same kind of cybersecurity authorization requirements (e.g.
username, password) as would be required through a typical vendor’s maintenance
tool. As of the beginning of 2022, the only aspect of device management discussed in
IEC 61850 is an ability to update IED rmware and con gurations, described in
Technical Report IEC TR 61850-90-16 [3].”
Access to the data within IEDs may be made for many purposes. Different users have
differing data access needs to perform their responsibilities. Work is underway to
de ne how an IED using IEC 61850 might impose role-based-access restrictions to
different data for different users, but this remains a work in progress and includes
considerable complexity. For non-IEC 61850 access, each different device may
impose whatever restrictions or controls the manufacturer implements through their
maintenance tools. In addition to this, the utility may determine that addition
cybersecurity access controls are required over and above those available natively
in the product tools. Enforcing the utility’s cybersecurity policies might require
additional procedures (work ows, training, etc.) or additional technical measures to
manage data access, create auditable logs, verify device operation, etc.
6. Other considerations
Following is a list of some factors that may affect design choices and therefore should
be considered during grid automation system projects:

Consider all applications or users that require access to information from the
system. Also consider potential future applications that might require such
access and how such future access might be provided.
Consider the functional requirements of subsystems:
Are there reasons why a particular technology may be mandatory or
preferred for some purpose?
Do subsystems need to share information with other subsystems? If so, how is
this done (technology, protocols, etc.) and how is it secured (veri ed,
authorized, monitored, etc.)?
Are there commercial considerations that might in uence or override
functional requirements?
Can various users / applications be given access to the same data? How is this
controlled (authorized, veri ed, logged for auditing, etc.)? Do users need to be
assigned to groups where access permissions are de ned for the group, or are
users individually assigned permissions?
Does information need to be protected (e.g. from accidental or unauthorized
alteration)? If so, from whom and how?
Can information available within a subsystem be mapped or translated into a
suitable format for use within another subsystem without loss of relevant data
(e.g. without loss of timestamp or without loss of semantics)? If such lossless
transformation is not possible, what limitations arise from this?
How will maintenance actions be performed (e.g. isolation of automation
functions when workers are present)?
What kinds of remote monitoring or control are allowed for operational purposes?
What kinds of remote monitoring or control are allowed for maintenance
purposes? What kinds of changes are permitted remotely? Has this policy been
reviewed to consider travel and contact restrictions such as may have been in
force during the SARS-CoV2 pandemic?
Is similar access required to all devices, irrespective of manufacturer, age,
protocol, etc., or are individual access methods for each device type appropriate?
Do some devices need common access and some need individual access? This
may require consideration of operational and maintenance functions, training
requirements, etc.
7. Conclusions
Utility automation systems consist of many differing subsystems. There is typically a
range of equipment types, functions performed, data available and communication
protocols supported by that equipment. Many users have a legitimate need to access
some subset of information available in the various subsystems and may have
differing needs for the timeliness of that access (from the immediate update of
operational SCADA data to historical archive data retrieval). An integrated
automation system should give all users timely access to all information they need to
perform their tasks, irrespective of the actual source of that information. All access
should be secured appropriately to allow authorized access and prevent
unauthorized access or alteration. Each user should have such access with the
minimum of impediment caused by cybersecurity functions that control the access.
The process should be suf ciently intuitive as not to require extensive training.
Any project to extend the automation system should be taken as an opportunity to

review all uses of data available in the new extensions and all requirements to
access the new extensions for engineering management of those extensions. Modern
utility automation equipment has a wide range of data available, extending beyond
that traditionally used for operational purposes. Providing appropriately secured

access to that information can enhance the capability and ef ciency of the utility
divisions that make use of that information. Throughout this process, the
cybersecurity aspects of access management need to be included in all
considerations.
Utilities perform many functions internally, with staff having a range of skills and
expertise. The automation systems they use must support those functions. A fully-
integrated automation system provides the synergy of information needed to
support the multi-skilled, multi-discipline workforce and enable the multiple
functions that they perform for grid reliability
References
1. National Institute of Standards and Technology: Special Publication (NIST SP) -
800-82 Rev 2 Guide to Industrial Control Systems (ICS) Security, NIST,
Gaithersburg, USA, 2015.
2. International Electrotechnical Commission: IEC TS 62443-1-1:2009 Industrial
communication networks - Network and system security - Part 1-1: Terminology,
concepts and models, IEC, Geneva, 2009.
3. International Electrotechnical Commission: IEC TR 61850-90-16:2021
Communication networks and systems in power utility automations - Part 90-16:
Requirements of system management for Smart Energy Automation, IEC, Geneva,
2021.
Contact Author: A.C. WEST 

CSE 024
Distinctive Downsizing of Cone-Type

Insulating Spacer for 245 kV Class GIS by
Functional Insulating Materials
AUTHORS
K. OKAMOTO - Fuji Electric Co., Ltd.

N. HAYAKAWA - Nagoya University
M. HIKITA - Kyushu Institute of Technology
H. OKUBO - Aichi Institute of Technology
K. KATO - National Institute of Technology, Niihama College
N. OSAWA - Kanazawa Institute of Technology
K. WATANABE - Nagase ChemteX Co.
K. ADACHI - Central Research Institute of Electric Power Industry
Summary
The downsizing of gas-insulated switchgear (GIS) and gas insulated transmission
line (GIL) will lead to a reduction in their cost, installation area, manufacturing
energy, and use of SF6, which is an extremely potent greenhouse gas. However, there
is a limitation in the insulation techniques by the conventional composite material
technology.
To begin with, in order to investigate the possibility of downsizing of a cone-type

insulating spacer of GIS, ashover voltage (FOV) characteristics by standard
lightning impulse (LI) voltage in the SF6 gas were studied by fabricating a 1/4 size
mini-model cone-type spacers applying nano-micro composite (NMC) materials
and/or permittivity (ε) functionally graded materials (ε-FGM).
For NMC materials, epoxy resin with SiO2 micro- ller (MC) was lled with 1 vol % of
nano ller (SiO2, TiO2, SrTiO3, Al2O3) to fabricate a 1/4 size cone-type spacer model. In
the case of positive LI voltage in 0.5 MPa-abs, the scale parameter (η) of the Weibull
distribution of the FOV of the nTiO2- lled and nSrTiO3- lled NMC materials was
increased by 12% and 20%, respectively, compared to the MC (without nano- ller).
For ε-FGM, 1/4 size cone type insulating spacer was also produced as well as NMC,
and FOV was measured. ε-FGMs with relative permittivities (εr) distributed between
12.7 and 4 were designed and fabricated using SrTiO3 and SiO2 micro- llers. The FOV

at 0.5 MPa-abs under a negative LI voltage of ε-FGM spacer is higher than that of a
uniform permittivity sample by approximately 23 %. Thus, it was clari ed, both NMC
and ε-FGM techniques can improve the insulating properties of the cone-type
insulating spacers, and especially, it was drastically for FGM.
Furthermore, by applying ε-FGM technique, an insulation spacer of actual the size

reduction by 30% in diameter compared with that used in conventional 245 kV class
GIS was achieved, and withstand LI voltage characteristics were veri ed. The
permittivity distribution of the ε-FGM spacer was optimized by an inverse
calculation technique with a newly developed electric eld analysis method. The
maximum electric eld stress in SF6 gas of FGM spacer is decreased to 0.74 a.u.,
compare to the stress (1.0 a.u.) at the same location of Uniform permittivity spacer,
and decreased to the same level as 0.77 a.u. on the coaxial HV conductor. Finally, the
FGM spacer with distributed permittivity of εr=10-4 using SrTiO3 and SiO2 llers was
fabricated. Under negative standard LI voltage with 0.5 MPa-abs, the FOV of FGM
spacer was to be 21-29 % higher than that of Uniform spacer with εr=4. In addition,
as a result of withstanding voltage test of LI ±1050 kV 15 times which was a type test
of standard LI voltage with 0.6 MPa-abs, of 245 kV class GIS regulated in IEC 606944,
there was no ashover, and the standard requirement was satis ed.
KEYWORDS
Insulating Spacer - GIS - nano-micro composite - functionally graded materials - flashover voltage
1. Introduction
The downsizing of gas-insulated switchgear (GIS) and gas insulated transmission line
(GIL) will lead to a reduction in their cost, installation area, manufacturing energy,
and use of SF6, which is an extremely potent greenhouse gas. However, there are
limitations to conventional composite materials technology. Therefore, Fuji Electric
Co., Ltd has participated in the "Development of Innovative Functional Insulating
Materials for Electric Power Apparatus" project subsidized by the New Energy and
Industrial Technology Development Organization (NEDO) since 2017, and it has
collaborated with several universities and materials manufacturers to use nano-
micro composite (NMC) material and permittivity functionally graded material (FGM)
to improve the performance of insulating spacers [1]-[7]. Fig. 1 shows an image of an
insulating bus comprising a cone-type insulating spacer and tanks. The target value
for downsizing this spacer is 30% of its diameter and 50% by volume.
First, in order to clarify whether NMC is applied to the cone type insulating spacer
and the insulating property is improved, a 1/4 size cone-type spacer model was
fabricated, and FOV in the SF6 gas was measured by standard lightning impulse
voltage.
Next, to clarify whether ε-FGM is applied to the cone type insulating spacer and the
insulating property is improved, similarly a 1/4 size cone-type spacer model was
fabricated, and the FOV was measured. And the inverse analysis system of
permittivity distribution was newly developed in order to form the electric eld
grading by ε-FGM accurately and ef ciently.

Finally, an actual size cone-type insulating spacer of 245 kV class GIS applied the ε-
FGM, was fabricated, and the FOV in the SF6 gas was measured by standard
lightning impulse voltage.
This paper describes the results and discussions of these studies.
Figure 1 - Concept of downsizing of a cone-type insulating spacer for GIS
2. Study with cone-type insulating spacer made

of nanocomposites
2.1. Experimental
Table 1 lists the nano- llers used in the
experiment. Bisphenol A epoxy resin
was used as the main agent, alicyclic
acid anhydride was used as the curing
agent, and dimethylbenzylamine Table 1 - List of nano llers for each sample
(BDMA) was used as the catalyst. The
micro- ller was lled with 47 vol% SiO2
(hereafter referred to as (a) MC). Four
kinds of nano- llers, SiO2, TiO2, SrTiO3,
and Al2O3, were selected and lled to 1
vol% of MC. Here, the above samples are
referred to as (b) NMC-nSiO2, (c) NMC-
nTiO2, (d) NMC-nSrTiO3, and (e) NMC-
nAl2O3, respectively.
Fig. 2 shows the cross-sectional

structure and a fabricated sample of a
1/4 size cone-type insulating spacer.
Each resin and catalyst was prepared by
dispersing the ller in a planetary
centrifugal mixer, casting them in a
mold, and heating and curing them in a
constant-temperature oven. The
structure of the center conductor was
Figure 2 - 1/4 size cone-type insulating spacer designed to suppress the electric eld
enhancement of the triple junction by
providing a shield for electric eld
relaxation inside and outside the spacer
on the center conductor side.

The sample was placed in a pressure vessel, evacuated, and sealed with SF6 gas to
0.5 MPa-abs. The negative and positive standard lightning impulse voltage (1.2/50
μs) was used as the test voltage. The voltage application was carried out by the step
by step voltage raising method. A standard lightning impulse voltage was applied in
300 kV to 20 kV steps, and the ashover generated voltage was measured as FOV.
The test was stopped when the superposition of the discharge path occurred.
2.2. Results and discussion

Fig. 3 shows the Weibull distribution of the FOV in SF6 gas at positive (black paint)
and negative (white blanking) impulse voltages. Table 2 lists the Weibull parameters
η and m . The table shows that the η of each NMC spacer was almost the same as that
of MC with negative impulse application, while the η of NMC-nTiO2 and NMC-nSrTiO3
was improved by 12% and 20%, respectively, compared to MC with positive impulse
application.
It has been reported that the initial electron emission from the insulating spacer
contributes to the polarity effect appearing in the impulse FOV of the epoxy spacer in
GIS. However, this study shows that the deep electron trap level is formed by
nanocomposite from the analysis by quantum chemical calculation, and that this
improves the FOV in the model electrode system in SF6 gas. From the above research
report, the effect of deep trap level formation by the nano ller addition on the
ashover voltage characteristics of the insulation spacer model obtained in this
study is discussed from the viewpoint of creeping discharge generation and
progress. That is, the following are assumed: FOV generation by the initial electron
supply from the insulating spacer surface and the streamer generation condition of
Schumann and the creeping discharge process from the streamer progress at the
insulator interface to the dielectric breakdown.
Table 2 - List of Weibull parameters under each

test condition for each sample
Figure 3 - Weibull plots of FOV by NMC in 1/4

size cone-type spacer model
Fig. 4(a) shows an example of the discharge trace after the ashover test, while Fig.
4(b) and 4(c) show a model diagram of the creeping discharge mechanism at negative
and positive impulse voltages for the insulating spacer model. As shown in Fig. 3 and
Table 2, the FOV of the nano- ller added spacer sample hardly changed in the
negative polarity, while the FOV of NMC-nTiO2 and NMC-nSrTiO3 greatly improved in
the positive polarity. That is, when the HV conductor had a negative polarity, the FOV
improvement effect was not observed by the addition of nano ller because the initial
electron injection from the HV conductor became the main factor of the FOV
determination. However, when the HV conductor had a positive polarity, the initial

electron emission from the deep trap formed on the surface of the insulating spacer
was suppressed to improve the FOV. The assumption of the initial electron
suppression was based on the deep trap level formation by the addition of
nanoparticles obtained from quantum chemical calculations [8].
Figure 4 - Example of discharge trace after flashover test and discharge model diagram of polarity effect
in GIS spacer model
3. Study with cone-type insulating spacer made

of functionally graded material
3.1. Design of ε-FGM by inverse analysis

The permittivity distribution of ε-FGM was manually determined using the results of
previous studies coupled with electromagnetic theory. For proper relaxation of
electric eld stress, many trials on the electric eld analysis for various kind of
permittivity distribution of ε-FGM were needed. To optimize the electric eld grading
by ε-FGM quickly and ef ciently, the inverse analysis of permittivity distribution was
used.
In this study, a cone-type FGM spacer made by variable compounding casting was
used. For this reason, it was assumed that the dielectric permittivity was in the axial
direction, and its distribution in the radial direction was uniform. Here, as shown in
Fig. 5(a), the inside of the spacer is divided into small regions, and each dielectric
permittivity is set to εi. The potential at the boundary of different regions on the
surface of the spacer is de ned as Vi.
Because each region is formed into a thin plate shape, it can be replaced with a
parallel plate capacitor. That is, it can be expressed by a series capacitance, as shown
in Fig. 5(b). To approximate the desired electric eld intensity distribution, the
desired dielectric permittivity distribution can be attained using an inverse solution
approach by considering a method to assign a voltage to each capacitance by
changing the dielectric permittivity. In the analysis, an appropriate dielectric
permittivity distribution was given to the spacer, and electric eld analysis was
performed using the nite element method to obtain the potential sharing of each
part of the spacer. From this potential sharing, the dielectric permittivity distribution
of each region of the spacer was attained and used to perform electric eld analysis.
This process was repeated to gradually reduce the electric eld intensity. Once the
electric eld intensity distribution converged, the calculation was terminated.

Figure 5 - Potential distribution around spacer

Fig. 6(a) shows the calculation model.
The ratio of the inner diameter to the
outer diameter of the coaxial cylindrical
conductor of the GIS was set to 1: 3. The
shield electrode was applied to the
center conductor of the cone-type
spacer at a spacer inclination angle of
45°. The inside of the spacer was
divided into small regions from # 1 to #
34, and the permittivity of each region
was obtained by inverse solution
calculation. An initial uniform
distribution of the relative dielectric
permittivity of 4.0 was given.
Fig. 6(b) shows the electric eld

intensity in and around the spacer
obtained as a result of the inverse
solution. When the dielectric Figure 6 - Calculation model and results (electric
permittivity in the spacer was uniform,
eld strength)
the electric eld reached its maximum
at the corner of shield electrode. On the
other hand, when the dielectric
permittivity distribution obtained by
the inverse solution was given, this
electric eld could be greatly reduced.
As described above, the maximum
electric eld around the spacer was
reduced by the inverse solution
calculation method of the dielectric
permittivity distribution, and as a
result, the electric eld value of the
in nite length bus without the spacer
(coaxial theoretical electric eld) could
be reduced [9].
3.2. Experimental
For ε-FGM, a 1/4 cone-type insulating spacer was fabricated by a exible mixture
casting method as shown in Fig. 7, and the FOV was measured. The ε-FGM spacer
had a seamless distribution of εr = 12.7 - 4.0 (red line) from the HV conductor to the
grounded tank in the radial direction. In comparison with the ε-FGM spacer, the
uniform spacer had a permittivity εr = 4.0. We used two kinds of llers, SrTiO3 and
SiO2, and their composite with epoxy resin. SrTiO3 had a large permittivity of εr=332
and, as the main ller, was used to determine the εr distribution of the ε-FGM spacer.
The mixing ratio of the two composites was continuously controlled with two syringe
pumps and a static mixer, and then combined to create a cone-type ε-FGM spacer
(diameter: 60 mm), as shown in Fig. 8. A uniform spacer with the same shape and
size as the ε-FGM spacer was fabricated using only the SiO2 composite.

Figure 8 - 1/4 size cone-type ε-FGM insulating
Figure 7 - Concept of flexible mixture method spacer
3.3. Results and discussions

The fabricated cone-type insulating spacer was placed in SF6 gas (0.5 MPa-abs), and
a standard lightning impulse voltage was applied in 300 kV to 20 kV steps, and the
ashover generated voltage was measured as FOV. The measured and estimated
values according to the volume-time theory are shown in Fig. 9. In the actual
measured values, the average values of 6 uniform and 14 uniform spacers for ε-FGM
and the FOV of the uniform spacer are in agreement with the estimated values.
Additionally, though the measured FOV of the ε-FGM spacer was higher than that of
the uniform spacer, the maximum value of the measured FOV is only slightly lower
than its estimated value. This is likely due to the production error in the dielectric
permittivity of the ε-FGM spacer. The maximum experimental value of the ε-FGM
spacer was veri ed to be 23% higher than that of the uniform spacer, which was also
consistent with the estimated value of the ashover voltage.
Moreover, the discharge of the uniform spacer is all creeping discharge on the
concave side, and it is suggested that the high-voltage shield edge, which has the
largest electric eld, is the origin. On the other hand, the discharge of the ε-FGM
spacer is gap discharge on the concave side (11 pieces) and a creeping discharge on
the convex side (3 pieces), suggesting that the dielectric permittivity distribution of
ε-FGM relaxes the electric eld at the shield edge of the high-voltage side and
improves the ashover voltage [10].

Figure 9 - Improvement of FOV by ε-FGM in 1/4 size cone-type spacer model
4. Study with actual size cone-type insulating

spacer
4.1. Design of cone-type

FGM spacer
For the ε-FGM, an actual size cone-type
insulating spacer of 245 kV class GIS
was fabricated, and the FOV was
measured. The permittivity distribution
of the ε-FGM spacer was optimized by
an inverse calculation technique with a
newly developed electric eld analysis
method.
Fig. 10 shows the cross section of Actual Figure 10 - Actual size cone-type spacer model
size cone-type FGM spacer. The outer
diameters of HV conductor is 90 mm
and the inner diameter of grounded
tank is 240 mm, respectively. The
con guration of the grounded tank
around the spacer is modi ed in
consideration of the actual GIS spacer.
The relative permittivity (εr) is reduced
in 6 steps from εr=10 around the HV
conductor to εr=4 around the grounded
tank.

Fig. 11(a) shows the electric eld
distribution around the Uniform spacer
with the same con guration as the FGM
spacer and a constant εr=4, whereas Fig.
11(b) shows the electric eld
distribution around the FGM spacer in
Fig. 10. The electric eld strength at the
shield edge of HV conductor in the
concave side of Uniform spacer is
de ned as 1.0 a.u., which is higher than
those on the coaxial HV conductor and
on the spacer surface in the concave and
Figure 11 - Electric eld distribution convex sides. On the other hand, the
electric eld strength at the same point
of FGM spacer is 0.74 a.u., which is 26 %
lower than that of Uniform spacer and
almost the same as 0.77 a.u. on the
coaxial HV conductor. Note that the
electric eld strength on the spacer
surface in the convex side is increased
to 0.73 a.u. from 0.50 a.u. of Uniform
spacer, which means that the electric
eld distribution in SF6 gas around the
FGM spacer is equalized owing to its
permittivity distribution.
4.2. Fabrication of actual size cone-type FGM spacer

As shown in Table 3, 2 kinds of composite materials are prepared in advance: a
SrTiO3 ller (εr=332) containing high permittivity composite material and a SiO2 ller
(εr=4) containing low permittivity composite material. By changing the ratio of these
two kinds of composite materials and mixing them, the permittivity was changed,
and by casting them into a mold sequentially, a cone-type FGM spacer with the
permittivity graded from εr=10 to εr=4 was produced.
Fig. 12(a) shows a uniform spacer of εr=4, and Fig. 12(b) shows a FGM spacer of
εr=10-4. In Fig. 12(b), the white region is εr=10, the black region is εr=4, and the gray
region between them is a graded region.
Filler material SrTiO 3 SiO 2
Mean diameter 1.0 – 1.5μm 1.5μm
Relative permittivity 332 4
Table 3 - Speci cations of ller materials

Figure 12 - Fabricated actual size cone-type spacer
4.3. Results and discussions

The fabricated cone-type spacer was installed in a tank lled with SF6 gas at 0.4-0.6
MPa-abs and exposed to negative standard lightning impulse voltage. The applied
peak voltage started at about 0.5 a.u., denoted by an arbitrary unit, and increased by
about 0.05 a.u. 3 samples of Uniform spacer and 2 samples of FGM spacers were
tested.
The cross symbols in Fig. 13 show the

measured ashover voltage of Uniform
(black) and FGM (red) spacers as a
function of SF6 gas. At 0.6 MPa-abs, 2
FGM spacers were ashover at 1.72 a.u.,
whereas 2 uniform spacers were
ashover at 1.4 a.u. and 1.5 a.u.
respectively. At 0.5 MPa-abs, one FGM
spacer was ashover in the range of
1.57 a.u to 1.67 a.u (acquired ve times)
whereas one uniform spacer was
ashover at 1.2 a.u. Thus, the FOV of
FGM spacers were experimentally
veri ed to have the higher by 17-20%
than the Uniform spacer at 0.5-0.6MPa- Figure 13 - FOV of cone-type FGM spacer
abs.
Fig. 13 also shows the estimated

ashover voltage, designated by circle
symbols (Uniform: black, FGM: red),
based on the Volume-time theory [6].
The dotted line is their approximations.
The measured FOV of FGM spacer
agreed well with the estimated ashover
voltage, respectively.

The ashover characteristics are
discussed in terms of ashover traces.
Fig. 14 shows the ashover traces of (a)
Uniform spacer at 0.6 MPa-abs, (b) FGM
spacer at 0.6 MPa-abs and (c) FGM
spacer at 0.4 MPa-abs, respectively. The
ashover traces of Uniform spacer in
Fig. 14(a) were found at the shield edge
of HV conductor in the concave side,
which corresponds to the point of high
electric eld strength in Fig. 11(a), as
well as on the spacer surface in the
concave side. On the other hand, as
shown in Fig. 14 (b) and Fig. 14 (c), there
were 2 types of patterns of the FGM
spacer: discharge from the concave-side
coaxial HV conductor to the ground
tank, and discharge from the coaxial HV
conductor shield edge to the spacer
surface near the ground tank.
These results suggest that the ashover

Figure 14 - Flashover traces of cone-type of Uniform spacer was a surface
spacers discharge originated from the shield
edge, whereas the ashover of FGM
spacer was a gap discharge between the
coaxial cylindrical electrodes.
Consequently, the fabricated cone-type
FGM spacer could exhibit its ashover
voltage improvement effect, compared
with the Uniform spacer, owing to the
electric eld grading and equalization
by FGM spacer.
In addition, a voltage withstanding test

of 15 times LI ±1050 kV, which is a type
test of the standard LI voltage of 245 kV
class GIS regulated in IEC 606944, was
carried out with the FGM spacer. As the
result, there was no ashover, and the
standard requirement was satis ed.
5. Conclusions
The summaries of this study are described below.
1. As a result of measuring the FOV in SF6 gas at 0.5 MPa-abs with 1/4 size cone-
type spacer applying NMC material, it was shown that, for positive impulse
voltages, the use of nTiO2 or nSrTiO3 increased the FOV by 12% or 20%,
respectively, compared with the MC.

2. By the inverse analysis method of the dielectric permittivity distribution, the
maximum electric eld around the ε-FGM spacer is reduced, and as a result, it is

possible to reduce the electric eld value to that of the in nite length bus without
the spacer.

3. FOV of the fabricated 1/4 size ε-FGM cone-type spacer was increased by 23 %
compared with that of the uniform spacer in SF6 gas at 0.5 MPa-abs under
negative standard lightning impulse voltage.

4. An actual-size ε-FGM cone-type spacer with a permittivity graded to 10-4 was
fabricated, and the FOV of negative lightning impulse voltage was measured. As a
result, it was veri ed that FOVs of this spacer can be improved by 17-20%
compared with a uniform spacer with εr = 4.
It was clari ed that both NMC and ε-FGM can improve the insulation characteristics
of the cone-type insulating spacer. And, by applying the FGM technology in the range
of pressure of 0.5-0.6 MPa-abs in 245 kV class GIS, it was clari ed in the actual size
that FOV of the FGM spacer increased about 21-29 % over uniform spacer of the
conventional technology. That is to say, this indicates that the application of the FGM
technology is very effective in gas insulation and gas solid insulation systems such
as GIS and GIL, and the target of size reduction of about 30% in comparison with the
conventional one at the beginning is expected.
In the future, AC withstand voltage test, partial discharge test, long term withstand
voltage test, etc. will be carried out for 30% downsizing of 245 kV class GIS spacer
and bus duct con gurations.
Acknowledgment
This paper is based on results obtained from a project (JPNP12004) commissioned
by the New Energy and Industrial Technology Development Organization (NEDO).
References
1. N. Matsuoka, Y. Fuchi, M. Kozako and M. Hikita, “Effect of Permittivity Variation on
Surface Flashover of GIS Epoxy Spacer Model in SF6 Gas,” IEEE ICD, pp.96-99,
2016.
2. K. Abe, K. Ohzuno, M. Kozako, M. Hikita, H. Mitsudome, H. Yanase and K. Okamoto,
“Partial Discharge Inception Voltage Characteristics for Nano-micro Composites
under Impulse Voltage in SF6 Gas”, IEEE CEIDP, pp.738-741, 2019.
3. Hitoshi Okubo, Hideki Shumiya, Masahiro Ito and Katsumi Kato,“Insulation
Performance of Permittivity GradedFGM (Functionally Graded Materials) in SF6
Gas under Lightning Impulse Conditions”, Conference Record of the 2006 IEEE
International Symposium on Electrical Insulation, IEEE, pp.332-335, 2006.
4. Hitoshi Okubo, Hideki Shumiya, Masahiro Ito and Katsumi Kato, “Optimization
Techniques on Permittivity Distribution in Permittivity Graded Solid
Insulators”, Conference Record of the 2006 IEEE International Symposium on
Electrical Insulation, IEEE, pp. 519-522, 2006.
5. H.Okubo, K.Kato, N.Hayakawa, M.Hanai, M.Takei, “Functionally Graded Materials
and their Applicationsto High Electric Field Power Equipment”, CIGRE SC D1 –
COLLOQUIUM IN HUNGARY BUDAPEST, 2009.
6. N. Hayakawa, K. Kato, H. Hama, Y. Hoshina and T. Rokunohe, “Electric Field
Grading Techniques in Power Apparatus Using Functional Materials,” CIGRE
Paris Session, D1-309, 2014.

7. N. Hayakawa, J. Ishiguro, H. Kojima, K. Kato, and H. Okubo, “Fabrication and
Simulation of Permittivity Graded Materials for Electric Field Grading of Gas
Insulated Power Apparatus,” IEEE Transactions on Dielectrics and Electrical
Insulation, Vol.23, No.1, pp.547-554, 2016.
8. Kazuma Abe, Kotaro Ohzuno, Masahiro Kozako, Masayuki Hikita, Hidetaka Masui,
Hiroshi Mitsudome, Hironori Yanase and Kenji Okamoto, “Flashover
Characteristics of Gas Insulated Switchgear Spacer Model using New Functional
Composite Insulating Material,” IEEJ 2019 Technical Meeting on “Dielectrics and
Electrical Insulation,” DEI-19-116, 2019.
9. K. Kato, H. Kojima, N. Hayakawa, H. Mitsudome, H. Yanase, K. Okamoto and H.
Okubo4, “Inverse Analysis of Permittivity Distribution of FGM (Functionally
Graded Materials) Insulator in Gaseous Insulation System,” ISH, 2019.
10. Y. Miyazaki, A. Izu, Z. Liang, H. Kojima, H. Masui, H. Mitsudome, H. Yanase, K.
Okamoto, K Watanabe, K. Kato, N. Hayakawa, “Breakdown Characteristics of Cone-
type ε-FGM Spacer for Gas Insulated Switchgears,” ISEIM, 2020.
Contact Author: K. OKAMOTO 

CSE 024
Synthetic Data – A Solution to Train

Diagnostic Systems for High-Voltage
Equipment without Fault-Condition
Measurements
AUTHORS
J.N. KAHLEN - RWTH Aachen University

A.MUHLBEIER - BatterieIngenieure GmbH
M. ANDRES, A. MOSER - RWTH Aachen University
B. RUSEK, D. UNGER, K. KLEINEKORTE - Amprion GmbH
Summary
D owntimes of high-voltage equipment can be reduced by detecting abnormalities
before faults occur, a task that can be performed by machine-learning (ML)
diagnostic systems. Training such diagnostic systems requires data containing fault
conditions. However, fault-condition measurements are often not available due to
low fault rates. This raises the question of whether ML-diagnostic systems trained
with synthetic data are valid predictors for real applications.
In this paper, synthetic data is generated with digital simulations of normal- and
fault conditions of a high-voltage pantograph disconnector to train multiple ML-
diagnostic systems. A system selection is executed using normal-condition
measurements and the synthetic data only. Finally, the performance of the selected
systems is evaluated with measurements of a disconnector with additional weight
load.
The results show that most of the ML-systems trained with synthetic data are
capable to detect the fault conditions. The systems based on arti cial neural
networks show the best diagnostic performance.
KEYWORDS
Fault detection - fault diagnosis - fault simulation - machine learning - power engineering
1. Introduction

To provide a safe, reliable and powerful electrical grid, a network operator has to take
on a range of duties and responsibilities. In Germany, such obligations are de ned
by the National Electricity Act (EnWG). According to §12 and §13 of the EnWG, the
network operator is obliged to provide a safe and reliable grid – both in the national
framework and where international tie lines are concerned [1].
The downtimes of high-voltage equipment can be reduced by using monitoring

systems that detect pre-fault conditions or signi cant deviations from the normal
conditions before the major fault occurrence [2]. For this purpose, different
monitoring systems are used to surveil key equipment on a continuous or event-
driven basis. They not only reduce the equipment’s downtime but also reduce the
amount of preventive maintenance. Such monitoring systems often base on special
user expertise. This approach has a limited ability to react on new unpredicted
conditions.
The digitalisation of energy systems in recent years has generated a large amount of
data that is dif cult to interpret by experts due to the numerous dependencies and
complexity of the systems. Diagnostic systems based on machine learning (ML) have
the potential to learn these complex dependencies [3] and diagnose pre-fault or fault
conditions [4]–[7]. The rules for evaluating the condition of the equipment are
determined by the ML algorithm [8]. To train the ML-based diagnostic systems well,
large amounts of data containing both normal and fault conditions are required [9]–
[13]. Due to the high reliability of equipment and low adoption levels of connected
monitoring systems, there are only limited measurements available, especially fault-
condition measurements [4], [14], [15]. A solution to this challenge is to train and
verify the ML-based diagnostic systems with synthetic data.
However, training ML-based diagnostic systems with synthetic data raise the
question of whether these systems can improve the quality of the diagnosis for real
applications.
This paper examines the training, veri cation and validation of diagnostic systems
trained with synthetic data, as shown in Fig. 1. A key in generation of proper training
data is to develop a computer-implemented electromechanical model of the device
to be diagnosed. Then, a statistical simulation method is developed to parametrize
the model for fault- and normal conditions. The basis for the fault-condition
simulations is the knowledge from operational experience and literature. Synthetic
data are generated for the drive motor current of a pantograph disconnector during a
making process and features are extracted from the synthetic data. In the second
step, ML-diagnostic systems and a threshold diagnostic system are trained and
created with the generated synthetic data of the pantograph disconnector. An
evaluation of the performance of diagnostic systems to determine the best diagnostic
systems for the task is carried out using the synthetic data and normal condition
measurements. Finally, the selected diagnostic systems trained with the synthetic
data are tested with measurements from a pantograph disconnector loaded with
additional weight. These measurements are generated by experiments in a
laboratory environment.

Figure 1 - Concept of the analysis
2. Disconnector Fundamentals: Components

and Faults
Air insulated disconnector’s main function is to create an electrical connection or a
separation between electrical circuits. They cannot switch load currents and but
create a visible disconnection. This allows safe working on downstream electrical
equipment [16]. Fig. 2 shows the basic structure of a pantograph disconnector. It
consists of a frame (1) holding the support insulator (2). The support insulator is
necessary as it carries the gear box (3) which is connected electrically and
mechanically to the disconnector arms (4) and is thus at high-voltage potential. The
mechanical force is generated by a DC-motor (5), transmitted by the rotary insulator
(6) and converted in the gear box into an up and down movement of the disconnector
arms. The arms must exert suf cient contact pressure to the contact point (7) to
ensure a good electrical connection [17]. A bad electrical connection could thermally
overload the disconnector during operation resulting in damages of the arms or the
counter contact of the power line.
Figure 2 - Structure of a pantograph disconnector

Various events or processes lead to major failures of air-insulated high-voltage
disconnectors with 0.29 major failures per 100 years per disconnector. The
frequency of these failures increases with the equipment ageing [18]. The failures
can be traced back to four main failure modes [19].
Current path: Increased resistance due to worn or corroded contacts,

Kinematic chain: Reduced material strength due to cracks of cast parts of the
disconnector or corrosion. This nally leads to braking of mechanical parts.
Drive: Corrosion due to water ingress into drive casing. Corrosion leads to
gearbox sluggishness or even a blocked motor.
Electrical control and auxiliary circuit: Corrosion leads to bad connections at the
auxiliary switches and other electrical components. Increased moisture also
causes short circuits of auxiliary contacts.
70.4 % of the disconnector major failures result in not closing or opening and 6.7 %
result in locking in the open or closed position [20]. Fig. 3 shows the origin of major
failures of 1,493 reported failures. The most major failures (34.8%) occur due to
failures in the mechanical operating mechanism.
Figure 3 - Origin of major failures of disconnectors [20]
The design of pantograph disconnectors in uences their reliability. Undersized

mechanical joints may result in faults in the kinematic chain due to material failure.
Insuf ciently sealed components, either due to design aws or due to material
failure of sealings, may result in corroded metal parts or in damaged electrical
components. Harsh weather conditions such as high salt content in the air may
accelerate these ageing processes.
3. Diagnostic System and Modelling

In this paper, three diagnostic methods are developed using with synthetic data. The
diagnostic systems are based on adaptive neuro-fuzzy interference systems
(ANFISs), arti cial neural networks (ANNs) and simple thresholds (Ts). ANFISs are
based on fuzzy logic. Fuzzy logic transforms a feature space into fuzzy sets. Rules
and membership functions imitate the continuous reasoning of humans [21] and can
be used for diagnostic purposes [22], [23]. ANFIS utilises neural learning methods to
optimize parameters and the structure of Fuzzy Interference Systems [24]. ANNs can
be used as a nonstatistical-based data-driven fault diagnosis tool and are well-
established [25]–[27]. The input of the diagnostic systems comprises features
extracted from continuous-time signals.

3.1. Generation of Synthetic Data Using Normal- and
Fault-Conditions Simulations
Often, only limited normal-condition measurements and no fault-condition
measurements are available. To circumvent this problem, synthetic data can be
generated using an electromechanical model. In measurements, a variation of the
measured variables can be observed due to external or internal in uences. These
in uences are often only partially represented in a model, as models typically
simplify the cause-effect relationships. It follows that the variation of the model
output does not match that of measurements. It is important that ML-based
diagnostic systems are trained with synthetic data that reproduce the variation of
measurements for robustness. Therefore, the simulation needs to account for the
variation in normal conditions into account that is not covered by the
electromechanical model. Furthermore, ML-based diagnostic systems need to be
trained with fault-condition data. As fault-condition measurements are often not
accessible, a fault-condition simulation needs to be implemented to generate
training data for supervised ML. The fault condition simulation also needs to account
for the variation that is not covered by the electromechanical model.
The simulation method consists of three core components: an electromechanical

model of the equipment, an algorithm for parameter optimisation of the
electromechanical model and the generation of parameter combinations of the
electromechanical model. With the parameter optimisation algorithm, the
electromechanical model can be tted to measurement data to minimise the error
between model output and measurements. With this algorithm and the available
measurements, a database of the parameters of the electromechanical model is
created by tting the electromechanical model multiple times to different
measurements. From this parameter database, another algorithm generates
parameter combinations for normal and fault conditions. With these parameters, the
electromechanical model is parameterised and simulated. Synthetic data can thus
be generated.
The simulation of the variation within the normal condition requires normal-
condition measurements of the monitored value or values to determine model
parameters. An optimisation of m model parameters is conducted to minimise the
error between the measurement and the model’s output. This results in n parameter
sets of 𝑆1,…,𝑛 for the model. Each set contains m parameters of 𝑃1,…,𝑚. These
parameter values are used to calculate probability distributions for each parameter
and to t a normal distribution, as shown in Fig. 4 a). When a normal-condition
simulation is conducted, the parameter values of the electromechanical model are
randomly selected according to the probability distribution of each parameter [28]–
[31].
For fault-condition simulation, it is necessary that fault conditions with a low

intensity or pre-fault conditions are simulated. This is because large deviations from
the normal condition can be identi ed using statistical or unsupervised learning
methods. For fault conditions with low intensity and pre-fault conditions, it is
assumed that the physical model is valid. For the simulation, it is not possible to
derive parameter distributions from measurements, as often no fault-condition
measurements are available. The fault-condition simulation creates new probability
distributions of selected parameters in regard to the previously derived parameter
distributions of the normal condition. The idea behind this is that if parameters are
selected outside of the established normal-condition distribution, a simulation with
such a parameter should result in an abnormal condition [28]–[31]. In Fig. 4 a) an

abnormal probability distribution is illustrated. The probability distribution ranges
from 𝑚𝑒𝑎𝑛𝑛𝑜𝑟𝑚𝑎𝑙 𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛+5 𝑠 to 𝑚𝑒𝑎𝑛𝑛𝑜𝑟𝑚𝑎𝑙 𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛+25 𝑠, where s is the standard
deviation of the normal condition’s normal distribution.
However, an abnormal condition is not necessarily a fault condition. For this purpose,
a fault de nition FD, measured in the standard deviation s, is introduced. Three fault
de nitions are shown in Fig. 4 b). If one of the model parameter values exceeds the
fault de nition, the simulation is considered as a fault-condition simulation. If no
parameter value exceeds the fault de nition, the simulation is considered as a
normal condition. Different fault de nitions are introduced in order to train
diagnostic systems with different sensitivities [28].
The introduced abnormal probability distribution ensures existence of normal-

condition data with a fault intensity close to the fault de nition. Depending on the
parameter value and the fault de nition, the generated synthetic data are labelled as
a normal condition or a fault condition.
The proposed method is not capable of simulating drastic fault conditions resulting
in a completely changed equipment behaviour. Only faults or failure modes that can
be linked to parameters of the electromechanical model can be simulated.
Figure 4 - (a) Normal- and abnormal-condition distribution of the motor friction parameter [28] / (b)
Fault de nitions [28]
3.2. Feature Extraction

To derive signi cant input data for the ML models, features are extracted from raw
signals. These features are the input for the diagnostic system. They represent
physical events in the monitored process or can be the result of mathematical
operations. Features can represent a variety of measurement values, such as mean,
mean square, standard deviation and integrals, as well as local minima or maxima
from the time domain, the derivative or the frequency domain.
3.3. Concept of the ML-Based Diagnostic Systems

ML models can be used as classi ers. To that end, ML models automatically learn
rules from given training data. The training data contain labelled data such as
‘normal condition’ and e.g. ‘fault type 1’. It is common to not provide raw signals as
input data for the ML models but extracted features from the signals. For the
diagnostic system, one model per simulated fault type is trained. The training data
are labelled such that only the data of the speci c fault type is labelled as faulty, the
rest is interpreted as ‘normal condition’. The model is trained with the labelled data.
After a successful training, the ML model is capable to predict whether unseen data
or extracted features belong to ‘normal condition’ or ‘fault condition’ on a scale
ranging from 0 to 1.
The structure of the complete diagnostic systems is shown in Fig. 5 and is divided
into four parts: data pre-processing, ML models, condition detection and fault
classi cation. In the pre-processing, features are extracted from the data. The ML
model part contains one ML model for each analysed fault type. Every ML model
outputs one value 𝑜𝑢𝑡1…𝑛. The outputs are combined to generate the diagnostic
systems’ output for fault detection as follows. If the output 𝑜𝑢𝑡1…𝑛 of any of the ML
models is greater than a threshold q, the diagnostic system’s output is ‘fault
condition’. If a fault condition is detected, the diagnostic system checks which ML
model’s output 𝑜𝑢𝑡1…𝑛 is the greatest. The fault corresponding to this particular ML
model is the output of the fault classi cation system.
Figure 5 - Concept of the ML-based diagnostic systems
3.4. Threshold System (T)

The advantage of the T-based systems is that they are quickly parameterized and the
results are comprehensible. The T-based system checks whether any feature 𝐹𝑛
exceeds a prede ned threshold value b, as shown in (1)[1]. O F is the output for each
n
check and b upper , F and b lower , F are the upper and lower thresholds. The output of
n n

the T-based system is determined by (2). If desirable, the thresholds can be adjusted
for sensitivity as per (3) and (4). 𝑏𝑢 0 is the initial threshold for the upper value and 𝑏𝑙
0 the initial threshold for the lower value. X is a factor for increasing (X > 1) or for
decreasing (X < 1) the threshold band.
⎧⎪
⎪⎪ 1 if F n ≥ b upper ,F ⎫⎮
⎮⎮
⎪⎪ n ⎮⎮
OF = ⎨⎪
⎪⎪ 1 if F n ≤ b lower ,F ⎮⎮
⎬⎮ (1)
n
n ⎪⎪ ⎮⎮
⎪⎪ 0 if b < F n < b upper ,F
⎮⎮
⎪⎪ lower ,Fn n
⎮⎮
⎩ ⎭
⎧⎪ normal condition , if max ⎧⎪ O ,. . .,O ⎫⎮ = 0
⎪⎪ ⎨⎪ F F n ⎬⎮
1
Out = ⎨⎪ (2)
⎪ ⎩ ⎭
⎪⎪ f ault condition , if max ⎪⎨ O ,. . .,O ⎮⎬ = 1
⎧ ⎫
⎪⎩ ⎪⎩ F 1 Fn ⎮
⎭
b u0 + b l0 b u 0 − b l0
b up , F = +X (3)
n 2 2
b u0 + b l0 b u 0 − b l0
b low , F = +X (4)
n 2 2
The T-based diagnostic system only detects but does not classify fault types. For a
fault type classi cation, a knowledge management system would be necessary.
[1] Note that the threshold for ML systems q is different from the T-based system threshold b.
3.5. Metric to Evaluate the Diagnostic System’s

Performance
Every ‘fault’ diagnosis can lead to expensive inspection or maintenance. Therefore, it
is important to pay special attention to the true negative diagnoses (equipment is in
normal condition and the diagnostic system has the output ‘normal condition’). The
weighted condition detection rate 𝑊𝐶𝐷𝑅𝑥 calculates the detection rate of a system
by weighting the diagnoses of the normal condition with a factor of a. The WCDR is
calculated using (13), where CD is the number of correct diagnoses and n the number
of normal conditions or fault conditions in the data set. The higher a, the bigger the
impact of true negative diagnoses on the WCDR is. Therefore, a should be chosen by
the equipment owner considering the individual inspection and maintenance
planning.
a CD normalCondition + CD faultCondition
WCDR a = (13)
a n normalCondition + n faultCondition
4. Results for a Pantograph Disconnector

The diagnostic systems for the pantograph disconnector detect faults by analysing
the motor current of the drive motor, as power consumption analysis is well suited
for fault detection and diagnosis [32]. Multiple ANFIS- and ANN-based diagnostic
systems with varying hyperparameters are trained with synthetic data with three
different fault de nitions FD. The generated diagnostic systems are broadly analysed
by examining the WCDR for normal-condition measurements and synthetic data. The
aim of this analysis is to select the optimal diagnostic systems for detecting faults of
a pantograph disconnector using only the synthetic data and available normal
condition measurements. The selected diagnostic systems are then tested using
measurements from a pantograph disconnector with additional weight load to check
if ML-diagnostic systems trained with synthetic data are valid
predictors for real-world applications.
4.1. Experimental Setup: Measurements in Normal and

Fault Condition
The validation of the diagnostic systems, requires real measurements. For this
purpose, experiments with a 245 kV pantograph disconnector are carried out. The
disconnector is switched multiple times and the current of the DC drive motor is
measured. The pantograph disconnector has to create an electrical connection with
the above located counter contact xed to the busbar, even if the height of the busbar
has changed e.g. due to a change in ambient temperature (see Fig. 6) The position of
the counter contact has a large in uence on the contact force and energy that has to
be provided by a drive. This variation is respected in the experimental setup by
varying the relative distance from the middle of the contact point of the disconnector
to the busbar between -230 mm and +230 mm (see Fig. 6). The ambient temperature
is kept in a range between 4°C and 8.8°C. With this setup, 61 measurements of a
disconnector in normal condition during a making process are be obtained for
further analysis.
As real fault-condition measurements are not available, the pantograph disconnector

is loaded with additional weight. Adding weights to the arms of the disconnector
imitate e.g. icing of the disconnector’s arms, strong corrosion of the joints or some
damages of the supporting spring (see Fig. 6). According to the utility experience, the
icing of arms and of the counter contact have led to a major malfunction of the device
in the past. Hence, this fault case is very relevant for the proper diagnostic. 4.5 kg,
13.6 kg, 18.1 kg, 22.6 kg are added sequentially to its arms (see Table VIII for the
corresponding quantity of measurements). The disconnector is operated with the
added weight and the DC drive motor current is measured.

Figure 6 - Schematic of the experimental setup
4.2. Extracted Features

In the case of the disconnector’s motor current during a making process, 51 features
are automatically extracted from the signals. Three classes of features (absolute
features, gradient features and deviation features) are shown in Fig. 7.
Figure 7 - Absolute (red) and gradient features (green) highlighted in the motor current signal. The
features represent physical events in the disconnector’s closing process

Feature Event
FA1t, FA1I Switching on the motor
FA2t, FA2I Inrush current of the motor
FA3t, FA3I No-load current with pre-resistor
FA4t, FA4I Current peak due to short-circuiting of the series

resistor
FA5t, FA5I No-load current without pre-resistor
FA6t ,FA6I Mechanical connection between motor and swing arm
FA7t, FA7I Begin of the upward movement of the arms
FA8t, FA8I Current peak during upward motion
FA9t, FA9I Arm contacts counter contact
FA10t, FA10I Maximal pressure against counter contact
FA11t, FA11I Re-activation of pre-resistor (maximum)
FA12t, FA12I Re-activation of pre-resistor (minimum)
FA13t, FA13I Switching off the motor (maximum)
FA14t, FA14I Switching off the motor (minimum)
Table 1 - FA features and the corresponding event during a making process

Feature Event
FG1t, FG1I Switching on the motor: maximal positive gradient
FG2t, FG2I Inrush current of the motor: maximal negative gradient
FG3t, FG3I No-load current with pre-resistor: maximal positive

gradient
FG4t, FG4I Mechanical connection between motor and swing arm:

maximal positive gradient
FG5t, FG5I Buildup of contact force: maximal positive gradient
FG6t, FG6I Re-activation of pre-resistor: maximal negative gradient
FG7t, FG7I Switching off the motor: maximal negative gradient
Table 2 - FG features and the corresponding event during a making process
4.3. Setup of the Experiment: Method for Training and

Evaluation of the Diagnostic Systems
Synthetic normal- and fault-condition training data are generated using the model of
a pantograph disconnector. The model is taken from the literature and implemented
in MATLAB Simulink [17]. The probability distributions of the parameters for the
simulation are identi ed using the 61 normal-condition measurements of the
disconnector’s motor current. The synthetic normal-condition data are veri ed by
means of available normal-condition measurements.
Seven physical and electrical faults, listed in Table III, are simulated. The faults are
based on the operational knowledge of utility personnel. For this paper, the
parameter distributions for simulating abnormal conditions are uniform and range
from 5 s to 25 s (Fig. 4 a). A starting point of 5 s ensures that the parameter to
simulate the fault condition is selected outside of the established normal-condition
distribution. Such a starting point results in fault conditions that match theoretical
fault analysis [28]. In total, 20,000 simulations are carried out in eight groups, each
containing 2,500 simulations. One group contains normal-condition data, where the
normal distribution of the parameter is assumed. The other seven groups contain the
synthetic data simulated with the abnormal conditions from Table III.

Fault Model parameter
Short circuit or deformation of the drive motor windings Factor for motor magnetisation function
Short circuit of the drive motor windings, increased Motor resistance value
cable resistance
Increased friction in the drive motor Coef cient of friction
Increased friction in the disconnector arm joints Coef cient of friction
Changed elasticity in the arm Spring constant
Inconstant supply voltage Voltage
Damaged drive motor current pre-resistor (reduced Resistance

resistance or shorted)
Table 3 - Physical and electrical faults simulated by parameter modi cation
In the next step, the main discrete features are extracted from every generated
synthetic signal. The ANFIS- [24] and ANN-based [33] systems are trained with the
features extracted from the 20 000 synthetic normal- and fault-condition data. The
task is to classify if the data belongs to a fault condition or a normal condition. For
the training, the simulations are divided into 80% training data and 20% validation
data for the training of the ANN- and ANFIS-based diagnostic systems to avoid
over tting.
To nd a diagnostic system with the greatest performance or correct diagnosis rate,

numerous ANFIS [24] and ANN models [33] are trained with different
hyperparameters (see Table IV and Table V accordingly). All combinations of the
hyperparameters are considered. ANNs with a linear node activation function act as
a linear regression model.

Parameter Investigated values
Training method - BFGS quasi-Newton backpropagation

- conjugate gradient backpropagation with Fletcher-
Reeves updates
- conjugate gradient backpropagation with Polak-Ribiére
updates
- gradient descent with momentum and adaptive
learning rate backpropagation
- Levenberg-Marquardt backpropagation
- one-step secant backpropagation
- resilient backpropagation
- scaled conjugate gradient backpropagation
- Bayesian regularisation backpropagation
Numbers of layers - one layer

- two layers
- three layers
Node activation function - hyperbolic tangent sigmoid

- linear function
Table 4 - Hyperparameters of the ANN systems
Parameter Investigated values
Training method - backpropagation

- hybrid method with backpropagation for the input
membership function parameters and least squares
estimation for the output membership functions
Numbers of considered indicators - three indicators

- ve indicators
- six indicators
Table 5 - Hyperparameters of the ANFIS systems
Because the computing time for the ANFIS training is remarkably longer than that
for the ANN training, a subset of the extracted features is used as ANFIS input. In
order to nd the most valuable features for each ANFIS model, a sensitivity analysis
for all features and the hyperparameters listed in Table IV is conducted. Each
training is repeated 24 times with the same con guration to minimize the possibility
that the nal result is a local minimum. To train diagnostic systems with different
fault sensitivities, three fault-condition de nitions (FD = 5 s, FD = 10 s, and FD = 15 s)
are determined, as shown in Fig. 4 b).

The initial boundaries 𝑏𝐼,𝑢0 and 𝑏𝐼,𝑙0 of the T-based system are set for each feature to
the minimal or the maximal feature value that appeared in the normal-condition
simulations. The factor X is varied in the range of 0 to 4 in increments of 0.1. The
trained diagnostic systems are analysed using the seven measurements from the
disconnector loaded with additional weight.
4.4. Evaluation of the Synthetic Normal-Condition Data

To evaluate the synthetic normal-condition data, 2,500 normal-condition data are
generated with the method described in section 3.1. The envelope of the synthetic
normal-condition data and the measurements are shown in Fig. 8. Features are
extracted from the measurements and the synthetic data. The probability
distributions of the features extracted from the synthetic data are compared to the
features of the measurements. This is to verify that the synthetic data model the
measurement suf ciently accurate. The boxplots of representative extracted
features, normalized to the mean of each feature of the measurements, are shown in
Fig. 9. The percental deviation of characteristic values of the synthetic data’s
boxplots to the corresponding boxplot values of the measurements are shown in
Table VI. The feature 𝐹𝐴8𝑡 , time of the local maximum during the upward
movement, has the largest deviation. The upper whisker of the synthetic data
deviates by 31.96% from the upper whisker of the measurements. As Fig. 8 shows,
the modelled upward movement between 5 s and 11 s does not show a distinct local
maximum as the measurements do. This results in big variations of 𝐹𝐴8𝑡 . The upper
whisker of 𝐹𝐴10𝐼 , the current peak during the build-up of the contact force, has a
deviation of 26.67 % from the measurements whereas the lower whisker has a
deviation of 0.01 % only. The mean of all the deviations of the characteristic boxplot
values is 5.57 % and the median of all features deviates by an average of 6.25 %.
This shows that the simulation is capable of modelling most of the distribution of the
features suf ciently accurate.
Figure 8 - Envelopes of the synthetic data and the measurements [28]

Figure 9 - Boxplots of selected features of the synthetic data and measurements. The data are
normalized to the mean indicator value of the measurements
Deviation from the measurements [%]
Lower whisker 25th- Median 75th- Upper whisker

percentile percentile
FA2I -13,66 -3,89 -0,85 1,37 9,68
FA4t 1,85 1,78 0,87 0,51 13,36
FA4I 2,74 9,95 11,01 0,11 1,14
FA8t 2,98 12,29 16,03 20,48 31,96
FA8I -22,92 -11,86 -10,83 -7,85 -2,50
FA10t -5,01 1,00 2,91 5,94 13,94
FA10I 0,01 11,31 13,92 12,33 26,67
FA13t -4,30 1,69 4,78 7,47 17,35
FA13I 14,06 17,39 18,41 14,45 12,45
Table 6 - Percental deviation of characteristic values of the synthetic data’s boxplots to the corresponding boxplot
values of the measurements
4.5. Evaluation of the ML Models

Fig. 10 a) shows the histogram of the output of the considered ANN models applied to
the simulated data. 99.7% of the ANN models converge either to one (fault condition)
or to zero (normal condition), with a tolerance of 0.02. Therefore, the fault

threshold 𝑞𝐴𝑁𝑁 of Fig. 10 can be set to any value between 0.02 and 0.98, and no
further investigation in nding an optimal threshold value is needed. For the
following evaluation, 𝑞𝐴𝑁𝑁 is set to 0.5.
Figure 10 - (a) Histogram of the ANN models’ output -(b) Histogram of the ANFIS models’ output
Fig. 10 b) shows a histogram of the output of all ANFIS-based models for all
simulated data. 92.1% of the outputs are in the range between –0.5 and 1.5. The
ANFIS models do not converge to certain values. The threshold 𝑞𝐴𝑁𝐹𝐼𝑆 needs to be
adjusted in order to nd the optimal fault thresholds. Therefore, 𝑞𝐴𝑁𝐹𝐼𝑆 is variated in
the range of –0.1 to 1.1 in increments of 0.1.
4.6. General Performance Evaluation of the Diagnostic

Systems
Multiple diagnostic systems with different hyperparameters are build. In a real-
world application, only could be implemented to detect faults. Therefore, a selection
of a diagnostic system needs to be conducted without using the measurements of the
disconnector loaded with additional weight. The diagnostic systems are evaluated
against measurements in different data sets:
measurements consisting of normal conditions only, as fault-condition

measurements are not available
simulated normal conditions
simulated fault conditions.
The results of all considered diagnostic systems (different hyperparameters such as

training methods, number of nodes, etc.) with a fault de nition of 𝐹𝐷 = {5 s, 10 s, 15
s} are presented in Fig. 11, Fig. 12 and Fig. 13. The fault intensity increases with an
increase in fault de nition. The selection of the best diagnostic system is a two-
criteria optimisation problem that is evaluated with Pareto frontiers (thick line).
Systems of the Pareto frontier describe the solution where no other system improves
one criterion without diminishing another criterion. A 𝑊𝐶𝐷𝑅20 is used to evaluate
the diagnostic systems for the simulated data. 𝑊𝑆𝐷𝑅1 represents the rate of correct
diagnosis for measurements where the weighting factor has no in uence.

Table VII lists the characteristic information of the Pareto frontiers for the
considered diagnostic systems:
the Pareto frontier tuple with the highest 𝑊𝐶𝐷𝑅𝑥 (HW)

the area under the Pareto frontier (AuF) (see also Fig 11.).
This characteristic information allows a comparison of different Pareto frontiers. The

AuF characterises the whole frontier whereas the HW characterises the optimal point
regarding one criterion. An AuF value close to 1 indicates the Pareto frontier of a
system with good performance.
Figure 11 - WCDR of the diagnostic systems for a fault de nition of 𝐹𝐷 = 5 𝑠.

For a fault de nition of 𝐹𝐷 = 5 𝑠 the Pareto frontiers of the T-based systems are
superior to the ANN- and ANFIS-based systems. The T-based systems have a smaller
HW for the simulations than the ANFIS- and ANN-based systems, but there are T-
based systems that always classify the measurements correctly, leading to the
largest AuF ( 𝐴𝑢𝐹 = 0.86 ) of the three methods. The results show that the ANN-based
systems are superior to the ANFIS- and T-based systems for a fault de nition of FD =
10 s and FD = 15 s, with AuFs of 0.966 and 0.998, respectively. For these fault
de nitions, there are T-, ANFIS- and ANN-based systems that always classify the
conditions of the measurements correctly.
The Pareto frontier systems of the ANN- and ANFIS-based systems show good
validation results for the normal-condition measurements, with a 𝑊𝐶𝐷𝑅1 of up to
100% for 𝐹𝐷 = 15 𝑠. However, there are ANN-based systems with good results for the
simulated data where the validation for measurements shows comparably poor
results (clustered in a 𝑊𝐶𝐷𝑅20 range between 99.1% and 99.8%, highlighted with a
red arrow). This indicates that ANFIS- and ANN-based systems trained with
simulated data also need to be validated with measurements.
An overall improvement in the Pareto frontiers with an increasing fault de nition can
be observed. This indicates that the transferability of the Pareto frontier systems
trained with simulated data to measurements is better with a fault de nition FD = 15
s. The ANN-based frontier systems outperform the T- and ANFIS-based ones.
Therefore, ANN-based systems, trained with synthetic data with a fault de nition FD
= 15 s and a high WCDR for measurements and simulated data should be chosen for
the real application.

FD T ANFIS ANN
5s HW for simulations (0.953 | 0) (0.980 | 0.097) (0.999 | 0)
HW for (0.794 | 1) (0.783 | 0.742) (0.809 | 0.661)

measurements
AuF 0,857 0,600 0,668
10 s HW for simulations (0.937 | 0) (0.972 | 0.11) (0.998 | 1)
HW for (0.860 | 1) (0.828 | 1) (0.995 | 1)

measurements
AuF 0,904 0,955 0,997
15 s HW for simulations (0.919 | 1) (0.923 | 1) (0.997 | 1)
HW for (0.924 | 0.548) (0.969 | 0.597) (0.998 | 0.823)

measurements
AuF 0,924 0,966 0,998
Table 7 - Characteristic information of the Pareto frontiers
4.7. Validation of Selected Diagnostic Systems

In section 4.6, it is shown that the ANN-based systems, trained with synthetic data
with a fault de nition FD = 15 s is most suitable for the real application.
Therefore, these diagnostic systems are evaluated for seven measurements of the
disconnector loaded with additional weight. Because the systems are not trained to
identify this additional weight load, the evaluation validates whether the diagnosis
output is plausible or not. The evaluation checks whether 1) all seven measurements
of additionally loaded disconnector are diagnosed as normal conditions (insensitive),
2) the rate of fault-condition diagnosis increases with increasing weight (plausible),
or 3) the rate of fault-condition diagnosis is unrelated to the weight (implausible).
Fig. 14 shows a detail of the results of the ANN-based systems trained with 𝐹𝐷 = 15 𝑠
and a high WCDR for both synthetic data and measurements. The ANN-based
systems are marked according to whether they are implausible, plausible or
insensitive. The gure shows, that none of the ANN-based systems trained with 𝐹𝐷 =
15 𝑠 and a high WCDR for synthetic data and measurements show implausible
diagnosis for the data of the disconnector loaded with additional weight. This
indicates that utilizing MLdiagnostic systems for real applications is possible.

The diagnostic system with the highest 𝑊𝐶𝐷𝑅20 for the synthetic data is insensitive
to weights. Therefore, the diagnostic systems, S1 and S2 showing plausible
classi cation results for the data of an additionally loaded disconnector are further
evaluated. Table VIII presents the detection rate for the systems S1 and S2 (marked
in Fig. 14) for the measurements from the disconnector loaded with additional
weight. Both systems are trained with a Levenberg–Marquardt algorithm and have
one layer with seven (System S1) and eight (System S2) nodes and a hyperbolic
tangent sigmoid transfer function. System S1 diagnoses all data sets with weights ≥
18.1 kg as fault conditions; System S2 diagnoses all data sets with weights ≥ 13.6 kg
as fault conditions.
Figure 14 - Diagnosis of additional weights on the disconnector arms for systems trained with 𝐹𝐷 = 15 𝑠
Diagnosis: Fault condition
Weight [kg] 4,5 13,6 18,1 22,6
Quantity of data 1 2 1 3
sets
System S1 0% 0% 100% 100%
System S2 0% 100% 100% 100%
Table 8 - Fault-condition diagnosis rate of systems S1 and S2 regarding weights on the disconnector arms
Additionally, the remaining ANN- and ANFIS-based diagnostic systems are tested
with the measurements from the disconnector loaded with additional weight. In total,
65.7% of the ANN-based systems trained with 𝐹𝐷=15 𝑠 are plausible, 15.0% are
insensitive, and 19.3% are implausible systems for weights attached to the

disconnector arms. For the ANFIS-based systems, 92.2% of the systems are
plausible, 7.7% are implausible, and 0% are insensitive when applied to the
measurements from the disconnector loaded with additional weight.
From the previous analysis we can conclude that transferring ML-based diagnostic
systems to real-world application is possible and even detect fault conditions they
are not trained for.
5. Conclusion
In this paper, it is examined whether synthetic data can be used to generate
diagnostic systems for real-world application when fault-condition measurements
are not available. ANN- and ANFIS-based diagnostic systems trained with synthetic
data and T-based diagnostic systems were considered. The synthetic data were
generated with an electromechanical model of a disconnector and a normal-
condition and a fault-condition simulation method. Different fault intensities were
considered in the process. A multi-criteria veri cation and validation was conducted
on the rate of correct diagnoses for measurements and simulated data to identify
potential diagnostic systems for real-world application. The multi-criteria evaluation
prevented selecting over tted of ANN and ANFIS models, models that always
predicting the normal conditions for measurements. The evaluation shows that the
rate of correct diagnosis for synthetic data depends on the fault type, the fault
intensity, the ML algorithm and the algorithm hyperparameters.
The ANN-based diagnostic systems with a high WCDR for both synthetic data and
measurements are the most appropriate for the pantograph disconnector. These
diagnostic systems are validated against measurements from a disconnector with
additional weight load to imitate a fault that the diagnostic systems are not trained
with. Even though the diagnostic systems are not trained for that speci c use case,
65.7% of the ANN-based diagnostic systems and 92.2% of the ANFIS-based ones
were capable of plausibly detecting the modi cation.
The investigation shows that ML-based diagnostic systems trained with synthetic
data are promising and it could be applied to other equipment and applications.
Further research is needed to validate the proposed method for other applications.
References
1. Ewg, Energiewirtschaftsgesetz: Gesetz über die Elektrizitäts- und Gasversorgung
EnWG. Letzte Änderung 18/12/2018., 2005.
2. Y. Han und Y. H. Song, “Condition monitoring techniques for electrical equipment
- a literature survey”, IEEE Trans. Power Delivery, 18. Jg., Nr. 1, pp. 4–13, 2003.
3. H. Su und K. T. Chong, “Induction Machine Condition Monitoring Using Neural
Network Modeling”, IEEE Trans. Ind. Electron., 54. Jg., Nr. 1, pp. 241–249, 2007.
4. M. Zekveld und G. P. Hancke, “Vibration Condition Monitoring using Machine
Learning” in IECON 2018 - 44th Annual Conference of the IEEE Industrial
Electronics Society, 2018.
5. S. Ma, M. Chen, J. Wu, Y. Wang, B. Jia und Y. Jiang, “High-Voltage Circuit Breaker
Fault Diagnosis Using a Hybrid Feature Transformation Approach Based on
Random Forest and Stacked Autoencoder”, IEEE Trans. Ind. Electron., 66. Jg., Nr.
12, pp. 9777–9788, 2019.
6. T. Boukra, A. Lebaroud und G. Clerc, “Statistical and Neural-Network Approaches
for the Classi cation of Induction Machine Faults Using the Ambiguity Plane

Representation”, IEEE Trans. Ind. Electron., 60. Jg., Nr. 9, pp. 4034–4042, 2013.
7. K.-C. Park, Y. Motai und J. R. Yoon, “Acoustic Fault Detection Technique for High-
Power Insulators”, IEEE Trans. Ind. Electron., 64. Jg., Nr. 12, pp. 9699–9708,
2017.
8. A. Singh und P. Verma, “A review of intelligent diagnostic methods for condition
assessment of insulation system in power transformers” in 2008 International
Conference on Condition Monitoring and Diagnosis: IEEE, 2008, pp. 1354–1357.
9. M. Dumitrescu, T. Munteanu, D. Floricau und A. P. Ulmeanu, “A Complex Fault-
Tolerant Power System Simulation” in 2005 2nd International Conference on
Electrical and Electronics Engineering: IEEE, Sep. 2005, pp. 267–272.
10. R. Gitzel, I. Amihai und M. S. Garcia Perez, “Towards Robust ML-Algorithms for
the Condition Monitoring of Switchgear” in 2019 First International Conference
on Societal Automation (SA), Piscataway, NJ, USA: IEEE, 2019, pp. 1–4.
11. Z. Meng, X. Guo, Z. Pan, D. Sun und S. Liu, “Data Segmentation and Augmentation
Methods Based on Raw Data Using Deep Neural Networks Approach for Rotating
Machinery Fault Diagnosis”, IEEE Access, 7. Jg., pp. 79510–79522, 2019.
12. Y. Wu, C. Lu, G. Wang, X. Peng, T. Liu und Y. Zhao, “Partial Discharge Data
Augmentation of High Voltage Cables based on the Variable Noise Superposition
and Generative Adversarial Network” in 2018 International Conference on Power
System Technology (POWERCON), Piscataway, NJ, USA: IEEE, 2018, pp. 3855–
3859.
13. S. Afzal, M. Maqsood, F. Nazir, U. Khan, F. Aadil, K. M. Awan, I. Mehmood und O.-Y.
Song, “A Data Augmentation-Based Framework to Handle Class Imbalance
Problem for Alzheimer’s Stage Detection”, IEEE Access, 7. Jg., pp. 115528–
115539, 2019.
14. A. Ypma, Learning methods for machine vibration analysis and health, Ph.D.
dissertation, Delft, Netherlands, 2001.
15. X. Li, W. Zhang, Q. Ding und J.-Q. Sun, “Intelligent rotating machinery fault
diagnosis based on deep learning using data augmentation”, J Intell Manuf, 31.
Jg., Nr. 2, pp. 433–452, 2020.
16. Handbuch Elektrotechnik: Grundlagen und Anwendungen für Elektrotechniker ;
mit 300 Tabellen, 5. Au ., Wiesbaden, Vieweg+Teubner Verlag / GWV
Fachverlage GmbH Wiesbaden, 2009.
17. B. Rusek, C. Neumann und N. Lambrecht, “Monitoring and Diagnosis of
Pantograph Disconnectors” in Internationaler ETG-Kongress, Vorträge des
internationalen ETG-Kongresses, J. W. Kreusel, Ed., Berlin, Germany: VDE-Verl.,
2009.
18. M. Runde, C. E. Sölver, A. Carvalho, M. L. Cormenzana, H. Furuta, W. Grieshaber, A.
Hyrczak, D. Kopejtkova, J. G. Krone, M. Kudoke, D. Makareinis, J. F. Martins, K.
Mestrovic, I. Ohno, J. Östlund, K.-Y. Park, J. Patel, C. Protze, J. Schmid, J. E. Skog, B.
Sweeney und F. Waite, “Final Report of the 2004 - 2007 International Enquiry on
Reliability of High Voltage Equipment: Part 1 - Summary and General Matters,
Working Group A3.06”, Nr. 509, 2012.
19. A. Maheswari, B. Rusek, B. Sweeney, C. Fischer, D. Makareinis, L. Gauthier, L.
O'Sullivan, N. Amyot, P. Del Crego Amo, T. Minagawa und T. Lindquist, “Ageing
High Voltage Substation Equipment and Possible Mitigation Techniques: Working
Group A3.29, Technical Brochure”, Nr. 725, 2018.
20. M. Runde, C. E. Sölver, A. Carvalho, M. L. Cormenzana, H. Furuta, W. Grieshaber, A.
Hyrczak, D. Kopejtkova, J. G. Krone, M. Kudoke, D. Makareinis, J. F. Martins, K.
Mestrovic, I. Ohno, J. Östlund, K.-Y. Park, J. Patel, C. Protze, J. Schmid, J. E. Skog, B.
Sweeney und F. Waite, “Final Report of the 2004 - 2007 International Enquiry on
Reliability of High Voltage Equipment: Part 3 – Disconnectors and Earthing
Switches, Working Group A3.06”, Nr. 511, 2012.
21. Z. Gao, C. Cecati und S. Ding, “A Survey of Fault Diagnosis and Fault-Tolerant
Techniques Part II: Fault Diagnosis with Knowledge-Based and Hybrid/Active

Approaches”, IEEE Trans. Ind. Electron., pp. 1, 2015.
22. S. Simani, S. Farsoni und P. Castaldi, “Fault Diagnosis of a Wind Turbine
Benchmark via Identi ed Fuzzy Models”, IEEE Trans. Ind. Electron., 62. Jg., Nr. 6,
pp. 3775–3782, 2015.
23. Q. T. Tran, S. D. Nguyen und T.-I. Seo, “Algorithm for Estimating Online Bearing
Fault Upon the Ability to Extract Meaningful Information From Big Data of
Intelligent Structures”, IEEE Trans. Ind. Electron., 66. Jg., Nr. 5, pp. 3804–3813,
2019.
24. A. T. Azar, “Adaptive Neuro-Fuzzy Systems” in Fuzzy Systems, A. Taher, Ed.,
Rijeka, Croatia: InTech, 2010, pp. 85–110.
25. Y. Shatnawi und M. Al-khassaweneh, “Fault Diagnosis in Internal Combustion
Engines Using Extension Neural Network”, IEEE Trans. Ind. Electron., 61. Jg., Nr.
3, pp. 1434–1443, 2014.
26. S. Toma, L. Capocchi und G.-A. Capolino, “Wound-Rotor Induction Generator
Inter-Turn Short-Circuits Diagnosis Using a New Digital Neural Network”, IEEE
Trans. Ind. Electron., 60. Jg., Nr. 9, pp. 4043–4052, 2013.
27. H. C. Cho, J. Knowles, M. S. Fadali und K. S. Lee, “Fault Detection and Isolation of
Induction Motors Using Recurrent Neural Networks and Dynamic Bayesian
Modeling”, IEEE Trans. Contr. Syst. Technol., 18. Jg., Nr. 2, pp. 430–437, 2010.
28. J. N. Kahlen, A. Mühlbeier, M. Andres, B. Rusek, D. Unger und K. Kleinekorte,
“Electrical Equipment Analysis and Diagnostics: Methods for Model
Parameterization, Fault and Normal Condition Simulation” in Proceedings of
International Conference on Condition Monitoring, Diagnosis and Maintenance,
Sep. 2019, pp. 43–52.
29. J. N. Kahlen, A. Würde, M. Andres und A. Moser, “Model-based Data Augmentation
to Improve the Performance of Machine-Learning Diagnostic Systems” in
Proceedings of the 22nd International Symposium on High Voltage Engineering,
Cham: Springer International Publishing, 2021.
30. J. N. Kahlen, M. Andres und A. Moser, “Improving Machine-Learning Diagnostics
with Model-Based Data Augmentation Showcased for a Transformer Fault”,
Energies, 14. Jg., Nr. 20, pp. 6816, 2021.
31. J. N. Kahlen, A. Würde, M. Andres und A. Moser, “Improving Machine Learning
Diagnostic Systems with Model-Based Data Augmentation - Part A: Data
Generation” in 2021 IEEE PES Innovative Smart Grid Technologies Europe (ISGT-
Europe): IEEE, Oct. 2021.
32. A. H. Sabry, F. H. Nordin, A. H. Sabry und M. Z. Abidin Ab Kadir, “Fault Detection
and Diagnosis of Industrial Robot Based on Power Consumption Modeling”, IEEE
Trans. Ind. Electron., 67. Jg., Nr. 9, pp. 7929–7940, 2020.
33. S. S. Haykin, Neural networks and learning machines, 3. Au ., New York, Pearson,
2009.
Biographies

Dr. Jannis Nikolas Kahlen was born in
Freising, Germany, in 1992. He received
his B.Sc., M.Sc., and Ph.D. in electrical
engineering in 2014, 2016, and 2021
from RWTH Aachen University, Aachen,
Germany. He works as a Research
Associate at the Institute for High
Voltage Equipment and Grids,
Digitalization and Power Economics,
RWTH Aachen University, and the
Fraunhofer Institute for Applied
Information Technology. His research
focuses on data analysis, arti cial
intelligence, simulation and modelling
for diagnostics and asset management
in the eld of electromechanical
devices.
Artur Mühlbeier was born in Almaty,

Kazakhstan, in 1987. He received the
B.Sc. and M.Sc. degrees in electrical
engineering from RWTH Aachen
University in 2010 and 2013
respectively. From 2013 to 2019, he has
been a research assistant at the
Institute for High Voltage Technology of
RWTH Aachen University. Since 2020,
he is with BatterieIngenieure GmbH in
Aachen, Germany.
Dr. Michael Andres, received his

Diploma Degrees in Electrical
Engineering (2011) as well as “Business
Administration & Engineering” (2012)
from RWTH Aachen University
(Germany). From then on he worked in
different leading positions for the
Institute of High Voltage Technology.
Michael received his doctorate in
Engineering (2016) from the RWTH
Aachen University gaining knowledge
on analysis and modelling of oil- lled
distribution transformers in normal and
overload operation. Since 2017, Michael
leads the department “Digital Energy”
in the Fraunhofer-Institute of Applied
Information Technology (FIT). The
department offers interdisciplinary
expertise from “Designing Future
Energy Supply Systems” up to “IT
Security Technologies”.

Dr. Bartosz Rusek received his M.Sc.
from TU Wroclaw in Poland and his PhD
from TU Darmstadt in Germany both in
electrical engineering. Since 2006 he
works for a German transmission
system operator Amprion GmbH in the
department of asset management. He
was involved the topics of EHV
equipment maintenance, renewal
strategies, life cycle models, asset data
management systems, dynamic line
rating and insulation coordination of HV
AC/DC hybrid lines. He is responsible for
integration of new innovative
technologies into EHV grid. He
contributes in different working groups
and steering committees in FNN, DKE,
CIGRE, Entso-E.
Dr. Dennis Unger was born in Lünen,

Germany, in 1985. He received his
diploma degree in electrical engineering
and management in 2009 and his Ph.D.
in electrical engineering in 2016, from
TU Dortmund University, Dortmund,
Germany. From 2015 to 2019 he worked
in the asset management department at
Amprion GmbH, Dortmund, Germany.
Since 2020 he is head of the department
Asset Management at Dortmunder
Energie- und Wasserversorgung GmbH,
Dortmund, Germany.

Dr. Klaus Kleinekorte received a
diploma in electrical engineering and a
PhD degree from the University of
Technology Aachen. Since 1992 he
worked with RWE in various
management functions. In October 2003
he was appointed to the Board of today’s
Amprion GmbH. As CTO his main area of
responsibility comprised Asset
Management/Grid Planning and the
Control Centre Brauweiler. As Chairman
of the UCTE Working Group Operations
and Security Mr. Kleinekorte was
strongly involved in the development of
the UCTE Operation Handbook. He was
member of the ENTSO-E Assembly.
Within CIGRE he represented Germany
at the Administration Council and the
Steering Committee.
Contact Author: J. N. KAHLEN 

CSE 024
Limits of electric eld for composite

insulators: state-of-the-art and recent
investigations of overhead line insulators
purchased by power utilities
AUTHORS
P. SIDENVALL, I. GUTMAN - I2G, Sweden

A. DECKWERTH - 50Hertz, Germany
P. DIAZ, P. MEYER - RTE, France
J.F. GOFFINET - Elia, Belgium
K. HALSAN - Statnett, Norway
M. LEONHARDSBERGER - APG, Austria
M. RADOSAVLJEVIC - Svk, Sweden
P. TRENZ - Bayernwerk, Germany
K. VARLI - Amprion, Germany
K. VALIMAA - Fingrid, Finland
Summary
Hydrophobic properties of composite insulators might be deteriorated in service
by continuous corona activity on the housing material itself or by corona from the
metal parts directed towards the housing. This issue should be considered already at
the design stage, i.e., the electric eld should be controlled and limited in a few
sensitive areas of an insulator. The criteria for limitation of the electric eld in the
sensitive areas of composite insulators are well established in the international
literature, but not implemented yet in CIGRE/IEC recommendations. In this paper,
the criteria are evaluated by electric eld calculations of twenty-six commercially
available insulators from seven different manufacturers. Laboratory test results then
con rm the results of calculation interpretation. It seems that manufacturers of line
composite insulators design their products with these criteria in mind. Thus,
standardization in relevant IEC standards should be considered, as these criteria are
already widely adopted.
composite insulator - electrical eld - hydrophobicity - water drop corona - laboratory test method

1. Introduction and goal
Historically, overhead line composite insulators were rst used in polluted areas to
reduce or eliminate pollution ashovers. Consequently, their dimensioning only
considered the pollution level and excluded possible in uence of ageing. A similar
situation with dimensioning was revealed after insulators started being installed in
relatively clean environments. Their hydrophobic properties were excellent when
they were new, but then the hydrophobic properties may be deteriorated in service
due to ageing by continuous corona activity on the housing material itself or by
corona from the metal parts directed towards the housing. This is a known ageing
mechanism speci c to composite insulators and should be considered at the design
stage of composite insulators together with the selection of grading/corona rings, i.e.,
the electric eld should be controlled and limited in three sensitive areas of an
insulator. To achieve a proper (optimal) dimensioning of composite insulators
equipped with grading/corona rings, three internationally recognized criteria in
critical areas should be considered [1], [2]:
1. The limit of electric eld on the grading/corona ring and end tting. Note, the
criterion cannot be properly applied for arcing horns because the electrical eld
might be intentionally increased on them to attract the power arc and protect the
hardware.
2. The limit of electric eld along the housing surface.
3. The limit of electric eld at the so-called triple point (where air and housing meet
the metal end tting). An example of the triple point adopted from a draft of the
revised IEC standard 62217 is shown in Figure 1.
Figure 1 - Illustration of the triple point at the sealing of composite insulator
All three criteria listed above are normally veri ed by electric eld calculations. The
rst criterion can also be veri ed by a standard RIV, or corona test described in IEC
[3] and [4]. The third criterion cannot be veri ed by a test but only by calculation.

The second criterion (the limit of the electric eld along the housing surface) is not
veri ed by any standard test, however, the Water Drop Induced Corona (WDIC) test
has been recently developed for this purpose [2], [5]-[8].
The goal of this paper is to update the state-of-the-art of criteria and test methods for
the veri cation of the limitations of the electric eld via literature and CIGRE/IEC
activities. Additionally, recent results of electric eld calculations and tests for the
commercially available insulators already purchased by different power utilities are
analyzed. Following the latter issue, a group of eight European power companies
(both TSOs and DSOs) together with the Independent Insulation Group (I2G, Sweden)
is running a collaborative research project intended to benchmark composite
insulators. The electric eld calculation/veri cation is one of the activities within the
project. The following companies supported this project: 50Hertz (Germany),
Amprion (Germany), APG (Austria), E.ON (Germany), Fingrid (Finland), RTE (France),
Statnett (Norway), Svenska kraftnät (Sweden).
2. State-of-the-art limitation of electric eld for

composite insulators
The data regarding the criteria for maximum electric eld for composite insulators is
very limited. According to CIGRE Technical Brochure (TB) 284 the levels of maximum
electric eld on the surface of composite insulators (at the tip of the rst shed from
the end tting) are estimated as 0,6-1,0 kV/mm [9]. It is, however, stated in the TB
that this estimation is too high to ensure a technically safe operation. Earlier EPRI
research indicated limiting a value of 0,45 kV/mm [10]. Earlier STRI research
indicated limiting at about 0,40 kV/mm [11]. This is valid for hydrophobic surfaces,
while continuous corona discharges can be found on hydrophilic surfaces at electric
eld levels as low as 0,05 kV/mm [11]. This was also observed in I2G/STRI tests. For
the maximum electric eld from the metal parts ( ttings and grading rings) CIGRE
TB [9] recommends the limit of 2,2 kV/mm. EPRI indicated 2,1 kV/mm [10], which was
often used as a reference value for design purposes. According to internal CIGRE
discussions, some utilities use values as low as 1,6 kV/mm. This is to consider for
possible non-uniformity of the surface of grading/corona rings during manufacturing
or contamination/ageing of the grading/corona rings in service. In the earlier paper,
STRI recommended 1,8 kV/mm [11].
Comprehensive research at STRI and EPRI summarizes the ndings independently
carried out by these two organizations to determine a practical limit for the
permissible electric eld on composite insulator surfaces for design purposes [1].
This paper published in 2015 covers investigations based on small- and full-scale
laboratory and eld tests performed by both organizations to re ne threshold levels.
Results from both natural ageing (STRI) and arti cial ageing (EPRI) tests were also
considered. The criteria derived from [1] have been used by STRI/I2G for several
years and many practical applications:
1. The limit of E- eld on the grading ring and end tting 1,8 kV/mm
2. The limit of average E- eld along the housing surface 0,42 kV/mm
3. The limit of E- eld at the so-called triple point 0,35 kV/mm.
More than 20 different designs of arrangements with composite insulators were

calculated to evaluate the limits of the electric eld and, in some cases, to make
recommendations for optimal design and positioning of grading rings and arcing

horns. Several results of calculations are presented in Figure 2 [2], [5], [6], [11], [12]
and some of them were also veri ed by service observations (normally performed by
daylight UV-camera), see Figure 3 [13].
Figure 2 - Typical examples of electric eld calculations for different designs of composite insulators
(adopted from [2]). Colour scale: 0 – 0,5 kV/mm tangential electric eld in top-right image and 0 – 2
kV/mm normal electric eld in remaining images. Blue colour indicates low electric eld while red
colour indicates high electric eld
Figure 3 - Left: tangential electric eld along insulator (dark red = above 0,5 kV/mm). Middle: inspection
by daylight UV camera revealing corona activity on insulator in humid service conditions. Right:
deterioration of the rst shed of the insulator due to corona (adopted from [13])
To verify the results of limitation of the electric eld along the housing based on
calculations, a proposal for the test was indicated in [1] and later the procedure for
the Water Drop Induced Corona (WDIC) test was developed and veri ed on several
insulator designs, both for single insulator strings and more advanced insulator
assemblies [2], [5]-[8], [14], [15], examples are shown in Figure 4. Comparison of
electric eld calculations and their veri cation by WDIC testing revealed that the test
method is reproducible (con rmed by a Round Robin Test in ve international
laboratories [8]), representative, repeatable, and cost-effective. Thus, the test ful ls
all four formal IEC requirements for the test procedure and will be included in CIGRE
TB developed by Working Group (WG) B2.57.

Figure 4 - Examples of insulators/insulated structures veri ed by WDIC tests. Top-left: 52 kV insulator
tested at the development stage of the test; top-right: full-scale 400 kV overhead line insulator;
bottom-left: full-scale insulated cross-arm 380 kV; bottom-right: full-scale three-phase “diamond”
insulation arrangement 420 kV
As the rst step for wider international acceptance, the proposed electric eld limits
and the WDIC test method were included in the draft of the TB prepared by CIGRE
WG B2.57. This WG is currently developing the guidelines for the application of
composite overhead line insulators and is expected to be published in the beginning
of 2022.
The need to control electric eld in critical (sensitive) areas of composite insulators
is also for the rst time described in the new draft of IEC 62217 (new Annex D) [16],
which is under consideration of IEC TC 36 MT 19. The criteria are also discussed in
the new draft of IEC 61109 [17], which is under consideration of IEC TC 36 MT 18.
3. Recent investigation of insulators from

storage of utilities
3.1. Insulators for the investigation

The complete list of insulators delivered by eight participating power utilities (i.e.,
already purchased commercial insulators) is presented in Table 1, where each
insulator is designated with a unique code from 1 to 26. Twenty-two (22) insulators
from this list were physically received for the WDIC test, while 26 drawings were
used for electric eld calculations. The test array was organized to cover many
different parameters:
Insulators made by seven different manufacturers (from “A” to “G”).

Different insulator designs such as standard and alternating shed pro les,
different design of ttings and sealings, etc.
Different years of manufacturing (2004-2020).
Different housing materials.
Different voltage classes, i.e., Um between 72,5 and 420 kV.

Different mechanical classes, i.e., SML between 100 and 500 kN.
Actual grading/corona rings and/or arcing horns according to provided drawings
(normally insulators 170 kV and up are equipped by grading rings).
Table 1 - Summary of insulators used for electric eld calculations and testing
3.2. Electric eld calculations

The electric eld calculations were performed for all insulators equipped with arcing
horns/grading rings recommended by manufacturers or purchased by power utilities
together with the insulators (detailed drawings were also provided), as intended in
service. All calculations were performed in Comsol Multiphysics 5.5 software [18].
The insulator strings were tted in a “standard” tower model most typical for each
utility, see several examples in Figure 5.

Figure 5 - Examples of different 400 kV “typical” tower models for the speci c power company used for
comparative calculations of electric eld and positions of investigated insulators marked by arrows
(corresponding in each case to the phase with the highest electric eld stress).
To simulate a worst-case scenario from an electric eld point of view, the 3D electric
eld calculations were performed for the phase having the highest electric eld
stress. Only six to eight sheds (or shed pairs in case of the alternating pro le) at the
HV end of the insulator were normally modelled. This simpli cation is based on I2G’s
earlier experience with similar calculations, where it was shown that only a few
sheds close to the insulator ttings are exposed to the highest electric eld stress [2].
Such simpli cation does not in uence the results but reduces drastic time for
calculations. The potentials of phase conductors and connected metal hardware were
set to values corresponding to the maximum system voltage. The potential of the
tower structure, ground surface, insulator upper (grounded) end tting, and shield
wires, were set to zero. If the arcing horn/grading ring arrangement was not
symmetric, the electric eld stress was evaluated on both sides of the insulator and
the highest values were conservatively used. The permittivity values used for the
insulating materials were as follows: air – 1; housing – 3 (for both silicone types, HTV
and LSR); glass ber core – 5. The chosen values are “typical” based on many years’
experiences and are applied to all types of silicone rubber insulators.
In general, the calculations covered the following:
the absolute electric eld stress on the arcing horn/grading ring,

the tangential electric eld stress along the insulator housing surface at the HV
end,
the tangential electric eld stress at the triple point.
Typical results of calculations are presented for the illustration in Figure 6.

Figure 6 - Examples of typical results obtained by electric eld calculations for two different insulators
(left-right). Top-left: normal electric eld stress at grading ring/arcing horn, colour scale 0-2 kV/mm
(from blue to red). Top-right: tangential electric eld stress along the housing used for the calculation
of average along 10 mm. Scale on y-axis 0-0,6 kV/mm. Bottom: close-up on top right image. Average
electric eld stress along 10 mm (red line) compared with the criterion 0,42 kV/mm (green line), scale
on y-axis 0,4-0,6 kV/mm
The results of the calculations are summarized in Table 2 using traf c light
principles (GREEN colour illustrates the results below the criterion and RED colour
illustrates the results above the criterion). For Emax values averaged along any 10
mm of housing surface, the results which are very close to the criterion of 0,42
kV/mm, are illustrated by YELLOW colour. The results of the electric eld calculations
allow to assume that all manufacturers are well aware about the proposed and
internationally disseminated criteria, despite that they are not yet prescribed by IEC.
This is because only 12% of the investigated insulators need small adjustments of
the position of grading rings to avoid water drop induced corona (normally 2-10 cm
up), and only 4% of the insulators (1 of 26) has too high electric eld stress at the
triple point. The excessive electric eld values noted on metal parts (grading rings
and ttings) are of second order importance in our case, because for all investigated
insulators possible corona from these parts will be oriented away from the insulating
body of the insulator.

Table 2 - Summary of results of electric eld calculations
Applying the proposed criteria for the limitation of electric eld stress, no grading
devices are needed for distribution class insulators (de ned as < 110 kV). For
transmission class insulators, one grading ring at the HV tting is normally needed
from 110 kV and up to 170 kV and two grading rings (at both ends) are required
starting typically from 220 kV. These estimations based on calculations are
con rmed by the answers to the specialized questionnaire distributed within this
project, see Figure 7. The answers to the questionnaire received from 50 utilities
cover the experience with about 9 million overhead line composite insulators
installed at transmission and distribution voltage levels. These insulators are
estimated to represent about 25% of all installations of line composite insulators
worldwide with the average time in service 24 years (with a maximum of 40 years)
[19]. One of the questions was if power utilities are using grading/corona rings and if
they do, at which voltage levels.
Figure 7 - Application of grading rings depending on voltage level (the results from the special
questionnaire distributed for power utilities). The Y-axis shows the number of answers related to the
speci c voltage level

3.3. Water drop induced corona (WDIC) test
Test method:
The test set-up and the test environment (minimum 45% humidity) were as per the
RIV-test (IEC 60437), with a simulated tower making the test arrangement
equivalent to service conditions. An example of the set-up is presented in Figure 8.
The wetting of the bottom part of the insulators was done by using the spray bottle
recommended for the wettability test according to IEC 62073, from a distance of
approximately 25 cm and lasting approximately 5 s, until the distribution of water
drops was saturated. All metallic parts were then properly dried with a piece of
paper/cloth to make sure that no water drop corona from metal would mask the main
test result, i.e., corona on housing. The conductivity of the water was as for standard
IEC rain, i.e., 100 μS/cm. The test voltage was applied in the same way as for the
corona test (IEC 61284), i.e., the voltage was rst increased to the corona inception
voltage (or, if reached rst, 120% of maximum operating voltage) and then reduced
to a target voltage of 100% of the maximum operating voltage. Observation of visible
corona was done at these two voltage levels, with the voltage kept for 60 seconds at
each level to allow for proper documentation using still image cameras.
The observation of corona was made by two standard photo cameras (Nikon D70),
installed in parallel, and using 30 seconds exposure time. One camera was placed on
a tripod on the oor and the other in the sky-lift at a height slightly above the bottom
of grading ring. It is important to document the results by photographs taken from
above and below the sensitive area of the test object since the corona activity could
otherwise be hidden from one of the directions.
The criterion was absence of visible corona at 100% of maximum operating voltage.
Examples of detected corona are presented in Figure 9 and Figure 10 and are
directly compared with the corresponding results of electric eld calculations.
Figure 8 - Example of set-up for the WDIC test for the cross-arm arrangement with simulation of both
conductor bundle and tower

Figure 9 - First example of typical results obtained in the WDIC test. Left: visible corona in the test;
right: clear indication to get visible corona in a sensitive area close to and at the sealing (respective
maximum electric eld >0,42 kV/mm)
Figure 10 - Second example of typical results obtained in the WDIC test. Left: visible corona in the test;
right: indication to get visible corona (respective maximum electric eld >0,42 kV/mm)
Test results
The results of the WDIC tests are summarized in Table 3. A Red-Yellow-Green traf c
light rating system is again applied, i.e., GREEN colour illustrates insulators which
PASSED, YELLOW colour illustrates borderline cases (valid only for electric eld
calculations), and RED colour illustrates that an insulator DID NOT PASS the test. The
results of tests are compared with electric eld calculations.

Table 3 - Summary of results of WDIC tests
A summary of the test results in comparison with the calculations is very promising
(considering note below Table 3):
For calculated values above the criterion of 0,42 kV/mm three out of three
electric eld calculations were con rmed by the test, i.e., 100%.
For calculated values below 0,42 kV/mm all fteen electric eld calculations were
veri ed by the test (no corona), i.e., 100%.
For calculated values very close to the criterion of 0,42 kV/mm (within 5%), three
out of four electric eld calculations were con rmed by the test, i.e., 75%. In this
speci c case, however, it might be complicated to be sure about the visibility of
corona.
4. Conclusion
The obtained results showed that the earlier proposed, internationally-disseminated
and recognized criteria for the limitation of the electric eld in sensitive areas of
overhead line composite insulators, work well. This was veri ed for the
comprehensive array of twenty-six commercial insulators from the storages of
participating utilities (26 to be calculated and 22 to be tested). The results of electric
eld calculations are also well supported by the results of the Water Drop Induced
Corona (WDIC) test. At present, these criteria are under discussion in two IEC
maintenance teams revising the IEC 62217 and IEC 61109. For IEC 62217, it has
already been decided to include a general description of the electric eld
requirements, leaving the criteria to the product standards. These criteria are
already included in several utilities' technical speci cations worldwide and it is high
time to nalize their standardization in relevant IEC standards.
The WDIC test complies with IEC requirements for a new test and became mature via
Round Robin Test in different laboratories. The comparison of test results with the
calculations showed the results are 100% con dent with the proposed criteria and
75% con dent in borderline cases very close to the criteria. This test will be included
in CIGRE technical brochure prepared by WG B2.57. This test can be easily

performed after the standard RIV test (using the same set-up) and may also be
included in technical speci cations of utilities as a complement for electric eld
calculations.
References
1. A.J. Philips, A.J. Maxwell, C.S. Engelbrecht, I. Gutman: “Electric Field Limits for the
Design of Grading Rings for Composite Line Insulators”, IEEE Transactions on
Power Delivery, Vol. 30, No. 3, June 2015, p.p. 1110-1118
2. I. Gutman, P. Sidenvall: “Optimal Dimensioning of Corona/Grading Rings for
Composite Insulators: Calculations & Veri cation by Testing”, World Congress &
Exhibition on Insulators, Arresters & Bushings, Munich, Germany, 18-21 October
2015
3. IEC Standard: “Radio interference test on high-voltage insulators”, IEC 60437
Second Edition, 1997-09
4. IEC Standard: “Overhead lines - Requirements and tests for ttings”, IEC 61284
Second Edition, 1997-09
5. P. Sidenvall, N. Sundin, I. Gutman, L. Carlshem, R. Kleveborn: “Development of
test method to verify composite insulators from water induced corona point of
view”, 32nd Electrical Insulation Conference (EIC), Philadelphia, Pennsylvania,
USA, 8 -11 June 2014, paper S11-3
6. P. Sidenvall, I. Gutman, L. Carlshem, J. Bartsch, R. Kleveborn: “Development of the
Water Drop Induced Corona WDIC Test Method for Composite Insulators”, IEEE
Electrical Insulation Magazine, November/December 2015, Vol. 31, No. 6, p.p. 43-
51
7. I. Gutman, A. Dernfalk: “Innovative testing techniques for veri cation of corona,
pollution and ice/snow performance of insulation structures”, World Congress &
Exhibition on Insulators, Arresters & Bushings, Barcelona-Sitges, Spain, 5-8
November 2017
8. P. Sidenvall, I. Gutman, L. Carlshem, J. Bartsch: “A Round Robin Test of the Water
Induced Corona Test”, ICOLIM-2017, Strasbourg, France, 26-28 April 2017, paper
0017
9. CIGRE WG B2.03: “Use of corona rings to control the electrical eld along
transmission line composite insulators”, Brochure 284, December 2005
10. A.J. Phillips, J. Kuffel, A. Baker, J. Burnham, A. Carreira, E. Cherney, W. Chisholm,
M. Farzaneh, R. Gemignani, A. Gillespie, T. Grisham, R. Hill, T. Saha, B. Vancia, J.
Yu: “Electric Fields on AC Composite Transmission Line Insulators”, IEEE
Transactions on Power Delivery, Vol. 23, No. 2, April 2008, p.p. 823-830
11. I. Gutman, J. Lundquist, V. Dubickas, L. Carlshem, R. Kleveborn: “Design of
corona/arcing rings when replacing cap-and-pin insulators by composite
insulators”, 17th ISH-2011, Hannover, Germany, 22-26 August, 2011, A-007
12. P. Sidenvall, I. Gutman, J. Schulte-Fischedick, J. Seifert, J-F. Gof net:
“Methodology of Modern E- eld Calculations - Case Study for Insulated Cross-
Arm”, CEIDP-2013, p.p. 334-337
13. M. Radosavljevic, I. Gutman, C. Ahlholm, P. Sidenvall: “Ageing and deterioration of
composite post insulators exposed to high electric eld in 220 kV and 400 kV
switchyards in Swedish network”, 2017 CIGRE SC B3 Colloquium, Recife, Brazil,
18-20 September 2017
14. P. Sidenvall, I. Gutman, J.-F. Gof net: “Application of new test procedure for
veri cation of water drop corona on innovative insulation cross-arms”, 19th ISH-
2015, Pilsen, Czech Republic, 23-28 August, 2015, paper 262
15. I. Gutman, J. Lundengård, S. Bucan, P. Sidenvall, J.-F. Gof net: “Trends in
pollution/corona testing for compact insulation systems in the form of insulated

cross-arms”, CIGRE SC D1 Colloquium, Rio de Janeiro, Brazil, 13-18th September
2015, paper 12
16. IEC 62217: “Polymeric HV insulators for indoor and outdoor use - General
de nitions, test methods and acceptance criteria”, Ed. 2.0, 2012-09 (under
revision)
17. IEC 61109: “Insulators for overhead lines – composite suspension and tension
insulators for a.c. systems with a nominal voltage greater than 1 000 V –
De nitions, test methods and acceptance criteria”, Ed. 2.0, 2008-05 (under
revision)
18. Comsol Multiphysics® 5.5, 2020, www.comsol.com.
19. I. Gutman, A. Deckwerth, K. Halsan, M. Leonhardsberger, P. Meyer, L. Diaz, M.
Radosavljevic, P. Trenz, K. Varli, K. Välimaa: “Application of Composite Insulators:
Perceptions vs. Service Experience”, 2022 INMR World Congress, Berlin,
Germany, 16-19 October 2022 (accepted for the publication)
Contact Author: I. GUTMAN 

CSE 024
Connecting two synchronous areas with an

HVDC link: The NordLink interconnector
AUTHORS
M. THIELE, M. VASOLD - TenneT TSO GmbH, Bayreuth, Germany

B.H. BAKKEN - Statnett SF, Oslo, Norway
Summary
NordLink is the rst HVDC interconnector between Norway and Germany. In this
paper we will explain how a dedicated Market & Operations (M&O) project for
NordLink was planned and organized in addition to the joint construction project.
Rather than working towards a common goal of constructing a new asset, the tasks of
the M&O project of NordLink can be viewed as building a bridge between the existing
market solutions and operational principles of the two TSOs. Key success factors
were the dedicated set-up of the M&O project, its early start and a structured and
faithful collaboration. In this way, all planned market solutions and operational
procedures were in place when the asset was ready for use. Thus, commercial
operation of NordLink could start as planned and has been going successfully to
date.
KEYWORDS
HVDC interconnector - Statnett - TenneT - Market&Operations - project management
1. The NordLink interconnector

NordLink is a bipole HVDC submarine electricity cable which connects Ertsmyra
substation in Tonstad, Norway and Wilster in Schleswig-Holstein, Germany. The 623
km long interconnector has a rated capacity of 1400 MW. The converters are voltage
source converters (VSC) with +/- 515 kV rated voltage. The interconnector started
operation in December 2020. The owners of NordLink are the Norwegian TSO
Statnett (50%), the Dutch and German TSO TenneT (25%), and KfW (25%), an
investment bank active in nancing energy transition projects, owned by the
German government.

Figure 1 - The NordLink interconnector
NordLink enables the exchange of renewable energy between Germany and Norway.
When Germany has surplus of wind and solar power this can be exported to Norway
while production from the Norwegian hydro power reservoirs is stopped or reduced.
Norwegian hydro and wind power can be exported to Germany when there is surplus
in Norway and/or low wind and solar generation in Germany.
Furthermore and independently from each other, the converters have the capability
to support the AC network with reactive power. This property is available from the
VSC converters independent of the amount of active power transmitted.
Alternatively, the converters can operate in the static synchronous compensator
(STATCOM) mode at zero active power. In this way, NordLink provides exible
reactive power control at the connection points of the converters.
Due to good planning and cooperation on all levels, the NordLink project was
completed on time and within budget in the middle of the Covid-19 pandemic. In this
paper, we will not elaborate on the construction of the interconnector but on the
market and operation parts of the project. In Section 2 we will explain how the
dedicated Market & Operations (M&O) Project for NordLink was planned and
organized. In Section 3 we will present the market solutions and operational
procedures that are currently implemented for NordLink. In Section 4 we will
present operational experiences to date, and nally, in Section 5 we conclude with
the main lessons learned and upcoming challenges.
2. Connecting two TSOs in two different

synchronous systems
There are already several HVDC interconnectors between the Nordic and the
Continental Europe (Core) synchronous systems, but NordLink is the rst
interconnector between the Norwegian and German power systems. Thus, the
project needed to build a bridge between the operational cultures and procedures of
the German TSO TenneT and the Norwegian TSO Statnett. Both Germany and Norway

are operating within the framework of the European guidelines for the Internal
Energy Market (IEM) [1] and of the Nordic/Core rules applicable also to existing
interconnectors. Even though the general framework is the same, there are
signi cant differences between Norway and Germany on a system operation level.
Furthermore, NordLink connects two different regulatory regimes for the rst time.
In addition to the main construction project teams, it was therefore necessary to
establish in 2016 a dedicated M&O Project responsible for operational agreements,
procedures and IT solutions. This project approach with a dedicated M&O project is
also new compared to previous interconnectors between Nordic and Continental
Europe systems.
2.1. The Market & Operations Project

Developing common solutions for market design and system operation between two
TSOs in two different synchronous systems is fundamentally different from
establishing a common project team with the goal of constructing a common asset.
Rather than working towards a common goal of constructing a new asset, the tasks of
the M&O Project of NordLink can be viewed as building a bridge between the existing
market solutions and operational principles of the two TSOs. This implies that all
market and operational solutions for NordLink have to t in seamlessly into the
existing landscape of procedures and market parties. Next to that, different
regulatory regimes on both sides of the border need to be respected and
international codes and guidelines have to be taken into account. The project was
structured with multiple phases in a top-down structure, as illustrated in Figure 2:
Figure 2 - NordLink M&O project phases
The rst phase is the Market & Systems Concepts phase where the teams come
together for the rst time to create a platform of common understanding and
joint purpose to agree on a high level concept for WHAT to use the link for. The
main deliverable of this phase is the System Operation Agreement (SOA),
describing the high-level principles of market solutions and operational
procedures including the responsibilities and liabilities of the parties with
respect to system operation.
In the second phase, called Market & Systems Processes, the teams go deeper
into the agreed concepts to agree on HOW to implement the various
functionalities of the link. There are two sets of deliverables from this phase: The
Functional Process Descriptions (FPD) and the Common and Internal Operational

Procedures (COPs and IOPs). The COPs and the IOPs are input to the operator's
handbooks and the later Operator Training phase. The FPDs are mainly input to
the third phase (ICT), but the COPs and the IOPs are also dependent on the
technical descriptions in the FPDs.
In the phase Information and Communication Technology (ICT) Solutions the
FPDs are used as a basis to establish detailed ICT speci cations and then
develop the necessary ICT solutions. Furthermore, extensive testing of
communication systems and IT functionality is performed. This is the most
resource demanding phase of the M&O project. The comprehensive IT testing
ranges from acceptance tests for individual IT solutions over integration tests of
interacting systems to end-to-end test along whole process chains between TSOs
and other parties. All this takes place without any physical exchange of energy.
The high-level IT architecture of NordLink is shown further below in Figure 5.
While the three phases described above focus on the operation of the link, there
is an additional phase/track called Operation & Maintenance which focus on the
handling of the asset itself. The main deliverable from this phase is the Operation
& Maintenance Agreement (OMA), describing the common procedures and
responsibilities for regular maintenance and repair preparedness for the asset.
Once the operational procedures, the ICT solutions and the asset maintenance
procedures are in place, the project organized a phase with extensive operator
training, both individually within each TSO and commonly between both TSO's
staff. The purpose of the common training was also to give operators the chance
to meet before the actual operation started, however, due to the Covid-19
restrictions during 2020 the common training could be done over video link only.
Naturally, there were a number of feedback loops between the phases. E. g. several
concepts that were agreed in the SOA had to be changed due to improved
understanding of common challenges and needs that were discovered during the
development of FPDs and COPs. Similarly, many of the FPDs had to be updated when
the ICT development started, both due to improvements discovered in the
implementation and due to new challenges that occurred.
The ve phases described above are the main content of the M&O project. In the nal
stages of the NordLink project, the deliverables from the M&O project and the
construction project come together in the Commissioning and Trial Operation Phases
as illustrated in Figure 3. These phases are de ned in the various construction
contracts to verify that the assets work as speci ed. However, these phases are also
important for the TSOs to verify that all the market systems, operational procedures
and IT solutions work as intended.
Figure 3 - NordLink commissioning phases (acronyms explained below)

2.2. Transmission Test Phase
In the Transmission Test Phase, a comprehensive set of technical tests are
performed. The starting point for the joint Transmission Test Plan is all the tests that
the contractor(s) need to verify asset functionality. In addition to these, the TSOs
have a number of tests they need to perform. Some of the TSO testing can be done as
an integrated part of the contractor tests, e. g. using the market system software to
schedule the detailed test pro les rather than running the tests manually. Other TSO
tests like Link Disturbance, Return to Service and TSO-TSO trading are not fully
covered by the contractor tests and need to be planned as additional tests.
During the Transmission Test Phase, the link is operating according to a planned
schedule. The TSOs buy and sell the necessary test energy in the energy market, but
the market does not in uence the volumes that are used. Only during high load
testing did the parties consider the price differences in the Norwegian and German
market, and test power was transmitted in the optimal direction from low to high
price areas when possible.
To manage this phase a dedicated test management plan and a concept for
procurement of test energy were established as a harmonized common guideline
between the TSOs. This included also detailed descriptions of roles and
responsibilities, a communication plan (weekly/daily meeting agenda) and
procedures for the actual test execution. A dedicated support team with extra staff
for the test energy procurement and a proactive information of stakeholders like
neighboring TSOs have greatly contributed to a successful test phase.
In parallel to the link testing, also the connections to external systems are tested in
this phase. This includes the Single Day-Ahead Coupling (SDAC), the Single Intraday
Coupling (SIDC) market XBID, Nominated Electricity Market Operators (NEMOs), the
Nordic Operations Information System (NOIS) and the Joint Allocation Of ce (JAO),
see chapter 3.1.
2.3. The Trial Operation Phase

When all components and functionalities are suf ciently tested and veri ed the link
can "go live", i. e. start commercial operation. In this phase, the contractors or the
TSOs do not any longer decide how to operate the link. Instead, the TSOs run their
normal capacity calculation processes to decide how much of the 1400 MW capacity
can be offered to the market on a daily basis. The market prices on each side of the
link will then decide how much, and in which direction, power will ow in each hour.
From the market player perspective there is no difference between the Trial
Operation and the Normal Operation phases. The difference is mainly in the
contractual structure, as the asset is still the responsibility of the contractors during
Trial Operation. The TSOs do not take over full responsibility and ownership until the
Trial Operation is successfully completed. The Operation & Maintenance Agreement
does not come into force until the TSOs have taken over the asset.
3. Overview of NordLink market and

operational solutions

3.1. NordLink market solutions
The European electricity market includes of a number of sequential markets and
products as illustrated in Figure 4. At present, NordLink participates in the European
day-ahead and intraday markets in line with current European and national
regulatory and legal requirements.
Figure 4 - European energy markets
Day-ahead market - NordLink participates in the European Single Day-Ahead

Coupling (SDAC) [2]
Available day-ahead cross-border capacity offered to the market is the
outcome of TenneT´s and Statnett´s daily capacity calculation process.
TenneT and Statnett currently use an NTC based approach. Flowbased
capacity calculation is being developed and will be introduced in the future.
In case of a decoupling situation in the day-ahead market, cross-zonal
capacities will be allocated in shadow auctions, which are performed by the
Joint Allocation Of ce (JAO) on behalf of European TSOs [3].
The embedded functionality for implicit loss handling in the day-ahead
auction is applied on NordLink [4]. This means that ow will only be
allocated on the interconnector if the value of trading power exceeds the
cost of the losses when transmitting the power. The traded power in the
sending end of the interconnector includes a linear approximation (market
coupling algorithm is currently only capable of processing a constant loss
factor) of the total amount of losses in cable and converters.

Intraday - NordLink participates in the Cross-border Intraday (XBID) market in
the European Single Intraday Coupling (SIDC) [5]
XBID offers continuous matching of trades across Europe, subject to available
cross-border capacity.
Un-allocated capacity from the day-ahead market will be made available to
the intraday market in accordance with intraday capacity calculations.
Cross-border intraday auctions (IDA) are planned to be introduced in Europe
in 2023.
Balancing services are currently not implemented on NordLink. The link shall
join the European platforms for manual and automatic balancing reserves (MARI
and PICASSO) when these become available between the Nordic and Core
synchronous systems [6], [7].
Countertrade / Redispatch is currently not implemented on NordLink.

Congestions in the surrounding AC grids are currently handled by other actions.
A future solution will as far as possible be harmonized with a common Nordic

methodology for countertrading and redispatch that is still in early development
phase.
Long-term transmissions rights are currently not implemented on NordLink. A

formal decision from the Norwegian and German National Regulatory Authorities
(NRAs) is needed before LTRs can be offered.
3.2. Deliverables of the NordLink M&O project

In this section, we brie y outline the main scope and deliverables of the NordLink
M&O project, i.e. all the deliverables that were necessary in addition to the direct
construction of the link.
3.2.1. Agreements
The M&O project delivered two main agreements:
The System Operation Agreement (SOA) that regulates all aspects of the use of
the link, including system operation governance, liabilities and responsibilities
between the parties. The SOA also takes into account the European regulations
and agreements that are applicable for the operations of the link both technically
and commercially. An important example for this framework is the ramping
limitations de ned for the two relevant synchronous systems Nordic and CE,
which NordLink must always comply with. The lowest value of the ramping
limitation given by the two systems will prevail.
The Operation and Maintenance Agreement (OMA) that regulates the operation
and maintenance of the asset itself, including maintenance, repairs, governance,
liabilities, insurance and general responsibilities between the parties.
3.2.2. Operational procedures and process descriptions
The day-to-day operation of the link is set out in 11 Common Operational Procedures
(COP) and further individual procedures at Statnett and TenneT. The operational
procedures are based on 8 Functional Process Descriptions (FPD) that go deeper into
the technical processes that are needed to support the operations, and thus serve
also as major input for ICT speci cations and development. Altogether the COP and
FPD describe the operational phases from outage planning over real-time operations
of the HVDC system and disturbance management to reporting and transparency:
Data Exchange of real-time measurements – FPD describes technical processes

to share information of each TSO’s SCADA system to the other to provide relevant
measurements within the respective observability areas.
Outage Planning – COP describes coordination of all planned outages that are
affecting the Net Transfer Capacity (NTC) of the link, including test activities and
work that causes an increased risk of tripping NordLink.
Switching procedure – COP describes procedures and responsibilities for
bringing NordLink into and out of operation.
Alternation of roles – COP and FPD describe operational and technical processes
of changing Leading and Following roles in TenneT's and Statnett's Planning
System, and Master and Slave for the HVDC control stations in the respective

Supervisory Control and Data Acquisition / Energy Management System
(SCADA/EMS).
Exchange of Capacity Data – COP and FPD describe operational and technical
procedures for determining and matching of NordLink capacity that is made
available to the Day-ahead and Intraday markets.
Energy Plan & Set Point Schedule – COP and FPD describe operational and
technical processes related to establishing and matching of the hourly Total
Energy Plan and the resulting Set Point Schedule, including error and deviation
handling.
Mutual AC-grid Assistance – COP describes operational procedures related to
TSO-TSO Exchange in different system states. This direct trade can be arranged
by phone on an hourly basis if required by operational situations emerging after
market closure.
Disturbance Management – COP describes operational procedure related to the
handling of unplanned events on the link.
Emergency Power Control (EPC) – COP describes the procedures for enabling and
disabling the EPC functions that are implemented on the link, manual EPC
activation and return to service after an EPC triggering. Triggering of EPC will
rapidly reduce the ow on the link, but not necessarily down to zero. The
triggering of EPC by either Statnett or TenneT takes place independently of
operational roles. FPD describes settings for the EPC functions that are
implemented on the link, and which events will lead to triggering of an EPC.
Reporting and Transparency – FPD describes technical processes regarding the
ful lment of European publication duties, limited to processes in which both
parties should participate actively or harmonize their practices. COP describes
how TenneT and Statnett coordinate and publish NordLink information to ful ll
the common transparency obligations, e. g. who is responsible for which
publication depending on the triggering event.
Exchange of Metered Values – FPD describes technical processes regarding
metering and exchange of meter values between the parties. These meter values
are used in the TSO-TSO settlement as well as in the balance settlement in each
country
Exchange of Calculated Values – COP and FPD describe procedure and technical
process for establishing the Daily Exchange Report (DER) containing e. g. values
of planned ow and imbalances on NordLink for a full day (24 hours). After
reconciliation and nal con rmation, the data is input to the TSO-TSO settlement
between Statnett and TenneT.
Operations Report – COP describes the structure of the Operations Report that is
used to inform the NordLink Steering Committee on the status and events related
to the operation of NordLink.
As the reactive power capabilities of both converters are independent from each
other and cause negligible losses in the converters, there is no need for common
procedures or processes on reactive power management. Both TSOs can use the
reactive power in control modes like reactive power control, QU-control and AC
voltage (droop) control. This is an important added value for AC grid operations in
the vicinity of the converter connection points.
3.2.3. IT solutions
A number of IT systems needed to be updated or expanded to operate the link,

mainly within the TSO environment but also external systems, see Figure 5:

Market & Scheduling systems (green)
SCADA/EMS systems (amber)
Settlement & Reporting systems (white)
External connections: Day-ahead/SDAC, intraday/SIDC, JAO, transparency etc.
Figure 5 - High-level IT architecture of NordLink
Naturally, the established IT architectures within TenneT and Statnett look different.
But within the parts relevant for NordLink the same hierarchy levels are visible:
The Market & Scheduling systems, which facilitate all market activities,
communicate with the external connections and give the required input to the
SCADA/EMS and Settlement & Reporting systems.
The SCADA/EMS systems, which control the operation of the link and give further
input to Settlement & Reporting. A crucial step here is to transfer the Set Point
Schedule matched between both parties to the power order sent to the HVDC
controller.
The Settlement & Reporting systems, which facilitate reporting and transparency
requirements and in the end also the nancial settlement of energy exchanged
via the link.
For the external systems, the required communication formats needed to be

implemented in the TSO systems. At the same time, necessary adaptations in each of
the external systems to incorporate NordLink needed to be ensured and agreed with
the responsible steering groups through formal change requests. In this way, a two-
TSO endeavor like NordLink is also dependent on projects and initiatives, which
involve many stakeholders and require multilateral agreements.
The particular challenge for the IT implementation was not to build one new stand-
alone solution but to adapt in parallel the existing IT systems embedded in the
established process landscape. This entailed an extensive coordination effort with
respect to speci cations of the interfaces, different IT providers involved,
simultaneous non-NordLink adaptions affecting the same systems, management of
test cases and timing of end-to-end tests. Nonetheless the whole IT landscape could
be tested successfully prior to the beginning of the transmission test phase.

4. Evaluation of operational experiences to
date
Commercial operations of NordLink started with the beginning of the Trial Operation
Phase on December 9th 2020. To allow for a secure start-up, several measures were
taken: First operations started at 9 am and not at the beginning of the business day
at midnight, so that control center shifts did have the support of both normal staff
and the project team. Furthermore, in a rst step only half the capacity was made
available to the market in order to check that all process run smoothly without risk of
a 1400 MW failure. For similar reasons the maximum ramping rates were reduced to
300 MW per hour.
These measures proved to be very valuable for a smooth transition into the
commercial operations and to gain rst operational experience. As a result, the full
link capacity could be released from January 5th, 2021. However, the capacity that is
offered on the German-Norwegian bidding zone border to the Day-ahead and
Intraday markets is always the lowest of three values: The technical capacity of the
link itself, the maximum export/import capacity in the German AC grid and the
maximum import/export capacity in the Norwegian AC grid. Events like planned and
unplanned outages in the connected AC grids may reduce a system's ability to both
import and export power in periods. Similarly, high levels of wind and solar
generation may also limit available capacity in the connected systems. These
capacities are calculated for all bidding zone borders in the common European
market each day and may result in an available market capacity that is lower than
the technical capacity of the link itself.
There may also be periods where the price difference between the Norwegian and
German bidding zones is too small to transmit energy pro tably (i. e. the value of the
price difference is less than the cost of losses on the link). Thus, the resulting ow
pattern as shown in Figure 6 features a high degree of volatility.
The ramping limitations for the Nordic HVDC interconnectors are based on common
Nordic stability criteria and agreed in the Nordic Synchronous Area Operational
Agreement. The purpose of these limitations is to reduce the deterministic frequency
deviations when several interconnectors need to change setpoint from one market
time unit to another. Up to now, a total Nordic limit has been shared equally between
all HVDC interconnectors with 600 MW per hour each, but with the introduction of
owbased market clearing this will become a dynamic limit to be allocated to the
border(s) with the highest social welfare gain.
It is worth noting there was a longer planned outage with further tests in March 2021
due to nal construction works at the end of the Trial Operation period. Apart from
some shorter outages, successful operation has continued to date.

Figure 6 - Operation of the NordLink interconnector from start of Trial Operation to November 2021
(weekly average flow in MWh/h, positive values correspond to flow from Norway to Germany)
5. Conclusion and outlook

In summary, the intensive and close collaboration within the NordLink Market &
Operations project has paid off well: All planned market solutions and operational
procedures were in place when the asset was ready for use. Thus, market operation
could start as planned and has been ongoing successfully to date.
Key factors for this success were the set-up of a dedicated M&O project, its early start
and a structured and faithful collaboration. In this way, it was possible in a timely
fashion to overcome the challenges of connecting different synchronous systems and
for TSO operations to act on eye-level to the construction work. Furthermore, it has
been of great value to involve the best process, ICT and HVDC-specialists of both
TSOs. Some resources where fully allocated to the project, while other key experts
were involved when needed. In particular, the involvement of these experienced
experts has ensured that NordLink solutions are not stand-alone but t into the
existing process landscape as good as possible. In this way, it was ensured that
adding new assets compatible to market and regulatory requirements was done
within a manageable effort for TSO control centers.
It cannot be overestimated that an M&O project approach for new “special”

challenges is of growing importance as the European energy system becomes more
complex, in terms of a dynamically expanding grid, new technologies being used,
many affected stakeholders and intricate ICT-tools involved. In addition to this, the
solutions need to be fully compliant with a complex set of common regulatory and
legal requirements governing the European Internal Energy Market. These aspects
shouldn’t be overlooked when addressing the bigger and costlier asset construction
tasks. At the end of the day, both parts are needed to develop the grids and markets
while ensuring security of supply.
Some further lessons learned include the importance of clarifying the roles and
responsibilities between the M&O project and the construction project. For future
projects it could also be helpful to improve the coordination between the project
phases developing the agreements and the FPDs/COPs to avoid some of the above-
mentioned feedback loops and enable an earlier start of the ICT phase.

Looking ahead, the further development of the NordLink processes is already
ongoing. A joint working group has been established to develop a common
Countertrade/Redispatch solution over the link.
There is also a common working group together with the Danish and Dutch TSOs
looking to implement new ramping rules for all HVDC interconnectors from Southern
Norway. Currently, there is an individual limit to how fast each interconnector can
ramp from one hour to the next, independent from what other interconnectors may
need to do. A more ef cient solution will be to implement an aggregated area limit
where one interconnector may ramp faster if the others do not need to ramp,
dependent on the price difference between the different bidding zones.
Furthermore, the implementation project of 15-minute products in the Nordics will

entail adjustments also on NordLink. Thus, the Statnett-TenneT collaboration on
NordLink will be a continuing exciting journey in order to exploit the full potential of
the interconnector within the dynamically changing European electricity grids and
markets.
References
1. https://www.europarl.europa.eu/factsheets/en/sheet/45/internal-energy-market,
https://www.entsoe.eu/network_codes/
2. https://www.entsoe.eu/network_codes/cacm/implementation/sdac/
3. https://www.jao.eu/
4. https://www.nordpoolgroup.com/trading/Day-ahead-trading/loss-functionality/
5. https://www.entsoe.eu/network_codes/cacm/implementation/sidc/
6. https://www.entsoe.eu/network_codes/eb/mari/
7. https://www.entsoe.eu/network_codes/eb/picasso/
Contact Author: M. THIELE 

CSE 024
Fast Frequency Response from

Transmission-Connected Solar Farms:
Australian Experience
AUTHORS
N. MODI, A. JALALI, I. COMMERFORD, A. GROOM

Australian Energy Market Operator (AEMO)
Summary
Increasing penetration of inverter-based resources (IBR) has expedited the
retirement of conventional synchronous generators (SGs) in power systems
worldwide. This has reduced levels of synchronous inertia and frequency control
provided by synchronous generators, which in turn increases the complexity of
primary and secondary frequency control. Synchronous inertia plays a key role in
slowing the rate of change of frequency (RoCoF) after a disturbance, providing time
for primary frequency control to act, and thereby maintain frequency within required
limits. Under low inertia conditions RoCoF can increase, reducing the available time
for primary frequency control to act.
Fast frequency response (FFR) can provide the necessary frequency control quickly
and can potentially help avoid undesired system frequency outcomes such as under-
frequency load shedding or over-frequency generation shedding. While to date
battery energy storage systems (BESSs) have been considered the most promising
technology to provide FFR, solar farms (SFs) have also demonstrated the capability
of providing FFR.
This paper present real-life example of FFR from SFs in the Australia’s National
Electricity Market (NEM) power system. Real-time measurements have been used to
benchmark a SF model, to con rm the accuracy of the model. The effect of varying
droop in the model has been assessed and its response is compared with response
from a BESS. This comparison demonstrates the capability of SF in providing FFR
that is similar to that from a BESS. The contribution of FFR from BESSs and SFs
toward overall power system security is demonstrated through a simulation of a NEM
region. The impact of droop and dead-band settings of SFs and BESSs on their FFR
provision, and their impact on power system security, are also assessed.

KEYWORDS
frequency control - synchronous inertia - rate of change of frequency - fast frequency response - solar
farm - battery energy storage - droop and dead-band
1. Introduction
The National Electricity Market (NEM) of Australia is experiencing an unprecedented
increase in the penetration of renewables and inverter-based resources (IBR) at both
transmission and distribution levels. The South Australian (SA) power system, in
particular, has witnessed operating periods with the instantaneous penetration of
IBR-generated electricity exceeding 140% of the region’s operational demand, with
the balance exported to neighbouring regions. The increasing level of IBR in the NEM
power system has accelerated the retirement of synchronous generators (SGs), which
has been a factor in reduced frequency control performance in the NEM [1]. The
Australian Energy Market Operator (AEMO) has been intervening to maintain a
minimum number of SGs online in several regions of the NEM, to maintain the
required level of system strength needed for stable operation of IBR [2]. While these
online SGs contribute to power system inertia, a steady decline in the inertia level
has been observed in the past few years[1].
Inertial response provided by SGs is an inherent and immediate release of the

energy stored in the rotary masses of SGs, opposing sudden changes in frequency
[3]. The inertial response of SGs is considered the main factor limiting the rate of
change of frequency (RoCoF) immediately after a disturbance. This provides
suf cient time for other frequency control measures to respond and control
frequency. Under low inertia conditions, limiting the size of the contingency has been
considered as an option for limiting the RoCoF, although the increasing sizes of
future IBR can increase the impact of this approach [1].
AEMO procure frequency control requirements through Frequency Control Ancillary

Services (FCAS) market. Two major categories of the frequency control requirements
are sourced through the market and they are regulation requirements and
contingency requirement. The current Frequency Operating Standards (FOS)
requires frequency to be regulated within ± 0.15 Hz during system normal operation.
This is achieved through procuring suf cient amount of regulation services through
the FCAS market. The Automatic Generation Control (AGC) uses procured amount to
regulate the frequency within the ± 0.15 Hz during system normal conditions. In the
event of a contingency different frequency operating requirements apply and is
dependent on the type of the contingency. If a contingency occurs and the frequency
leaves the normal operating frequency band (± 0.15 Hz), the dispatched resources
are required to provide suf cient amount of frequency control to control the
frequency as per the FOS. In 2020, the Australian Energy Market Commission
(AEMC), which is the rule making body in Australian power grid, re-established
effective frequency control within the normal operating frequency band by
introducing narrow dead-band frequency responsiveness from generators including
IBRs starting from no more than 0.015 Hz away from the nominal frequency. This
change does not ensure reservation of headroom unless the resource is dispatched
to provide frequency control.
Fast frequency response (FFR) has been gaining increasing attention to support
power system operation with low inertia, assisting in limiting the RoCoF, and
providing suf cient time for primary frequency control. Here, FFR acts as a rapid

active power increase or decrease, by generator or load, in sub-second timeframes to
correct frequency changes [4]. While FFR might not be able to fully replace the
physical inertia, it allows for larger contingency sizes at lower inertia levels, as is
demonstrated in this paper.
The NEM has one of the highest penetrations of distributed photovoltaic (DPV)
systems worldwide, causing signi cant operational demand reduction during sunny
periods [5]. Output from this DPV cannot be readily curtailed. Under low demand
conditions driven by DPV, and particularly for an islanded region, output from
transmission-connected SFs (and wind farms [WFs]) may have to be curtailed
instead, due to both insuf cient demand and as a result of other technical network
constraints [2]. This creates headroom, providing an opportunity for SFs to provide
FFR to the grid during these periods. Instances of relatively fast generation change
to correct frequency deviation by SFs have been observed in some real-life operating
conditions in the NEM power system, which is discussed later in this paper.
Worldwide, FFR has been assessed from various angles:
FFR has been investigated [6] from an energy viewpoint, where the energy
delivered after a frequency event is decoupled, based on the timeframe the
energy is released into the grid.
The capability of FFR supplied by a BESS, in replacing SGs’ inertia and primary
frequency response (PFR), has been evaluated [7], measuring the energy
provided by SGs to estimate the required size of BESS to deliver the same
amount of energy during the rst 6 seconds after the event.
FFR has been contrasted against synthetic inertia, where the former is de ned as
any type of fast active power change based on frequency deviation, while the
latter is the mimicked response of SGs’ rotational masses which is in proportion
to the RoCoF [8]. The reference shows that FFR leads to a better overall frequency
performance compared to synthetic inertia.
An SF has been modelled to simultaneously provide FFR and oscillations
damping to the grid [9]. The proposed model is evaluated on a two-area test
power system.
The frequency support capability of a curtailed grid-scale SF, among its other grid
support services, has been tested by the National Renewable Energy laboratory
(NREL), as reported in [10] and [11]. A ramp-rate limit of 10 MW/sec, however, is
applied in [10] to the SF’s response which could prevent the SF from delivering
its full response before the frequency nadir/peak and hence contributing to
arresting the frequency nadir/peak.
To date not much study has been carried out focusing on FFR provision from grid-
scale SFs and BESSs and their impact on the frequency stability of an actual large-
scale power system, supported by real-life observations. Also, the extent to which the
droop and dead-band settings of FFR providers can impact the frequency
performance of the grid has not been paid due attention in the past.
This paper rst presents the real-life responses from an SF and a BESS in the NEM
power system during frequency events. The measured responses of the SF and BESS
during the events are then replicated through electromagnetic transient (EMT)-type
simulations, via injecting the frequency signals measured during the events into the
EMT model of a SF and BESS. Through this exercise it is demonstrated that properly
con gured SF can provide FFR similar to BESS. A detailed wide area EMT model of a
gigawatt-scale NEM region is then used to demonstrate the potential bene ts of FFR
provision by SFs in addition to BESSs for frequency control of the grid. Additional

scenarios are also studied to evaluate the impact of droop and frequency dead-band
settings of SFs and BESSs on their FFR provision capability and impact on overall
power system security.
2. Real-life Frequency Response from SF and

BESS
There have been a few recent frequency events in the NEM for which AEMO has
received high resolution data from online generators. This data con rmed that grid-
scale BESSs responded to the frequency events by changing their active power
rapidly, consistent with their control design.
The received data also con rmed that some SFs in the NEM had provided similar
rapid change in the generation to control the frequency change, again in line with
their control design. In this section, examples of real-life response provided by a
transmission-connected BESS and a SF during a few frequency events are presented.
2.1. FFR from SF

Figure 1(a) shows the response of the SF to an under-frequency event where the
frequency dropped to 49.77 Hz. At the time of the event, the SF output was partially
curtailed, and SF was generating 50 MW prior to the event. If the SF was not
curtailed, it would have been generating 80 MW. Because of the curtailment, the SF
had headroom available to increase output. While the SF was operating with some
headroom, a frequency event occurred, and as shown in Figure 1(a) the output of the
SF started to increase with some time delay. The small time delay in the response is
associated with the frequency dead-band, and communication and processing time
delays in the controls, which can add up to 200 ms or more.
Figure 1(b) shows the response of the same SF to an over-frequency event, where the
frequency increased to around 51.1 Hz. As seen, the SF rapidly reduced its output
from around 90 MW to 52 MW, which is well aligned with its known droop (5%) and
dead-band (± 0.15 Hz) settings. The rapid reduction in the generation occurred prior
to the peak frequency, and therefore contributed towards arresting the over-
frequency.
Figure 1 - Measured SF responses to (a) an under-frequency event and (b) an over-frequency event

A noteworthy point in the SF response seen in Figure 1(a) is that after the initial
active power rise by the SF, in response to the frequency drop, the SF shows some
active power oscillations between t=8 and t=12 seconds and the active power
oscillations are somewhat in phase with the frequency. This can be associated with
the inherent delayed response from the SF, considering the frequency dead-band,
ramp rate, and the time required for measurement, processing, activation and
communication. However, given the very small size of SF’s active power oscillations
(less than 1 MW) and the Gigawatt scale of the network, it is unlikely that the
frequency oscillations are created by the SF response. Nevertheless, the delay in the
response of the SF to the frequency changes could impact overall frequency
performance of the grid. Hence, a careful design is required to avoid such a delay
and its subsequent consequences.
Another point to be noticed in Figure 1(a) is that the response provided by the SF is
as per its 5% droop setting. Therefore, even though larger headroom was available
for the SF during the event, only a small portion of it was used in line with its droop
setting. It is, however, observed that the SF has withdrawn its FFR after t=12s, while
frequency has not yet returned within the ± 0.15 Hz dead-band. The sooner-than-
expected FFR withdrawal could be associated with the fast variations in the
irradiance level observed over the time of the event, given the small size of active
power change (less than 3 MW). It could also be caused by the hysteresis-like dead-
band re-entering limit in the control system of the SF.
2.2. FFR from BESS

Figure 2(a) shows the response of the BESS to an under-frequency event, where the
frequency dropped to 49.62 Hz. As seen, the BESS has responded to the frequency
deviations quickly by changing its output power according to its known droop and
dead-band settings.
Figure 2(b) shows the BESS response to the same over-frequency event as in Figure
1(b), where the frequency rose to 51.1 Hz. As seen, the BESS has responded to the
event by reducing its output to -87 MW in less than 2 seconds according to its droop
(1.7%) and dead-band (± 0.15 Hz) settings.
Figure 2 - Measured BESS responses to (a) an under-frequency event and (b) and over-frequency event

The noisier frequency signals in Figure 2(b) could be due to not applying adequate
ltering to the measured signal. Also, it is observed that the update rate of the
frequency measurement in Figure 2(a) is around 200 ms, which is higher than that of
the other measured frequency signals, although it is suf ciently small to show the
overall trend of frequency changes.
3. Validation of Solar Farm model

The purpose of this section is to evaluate the accuracy of the EMT model of the SF in
emulating the behaviour of the actual plant. For this purpose, the frequency and
voltage signals measured at the terminal of the SF, for each event, were applied at
the point of connection (PoC) of the SF’s EMT model, using a playback set-up as
shown in Figure 3. The active power response of the SF model to the injected signals
was then compared against the measured active power response of the SF.
Figure 3 - Playback set-up to simulate the FFR of SFs
3.1. Responses to Under-Frequency and Over-Frequency

Events
The active power response from the EMT model of the SF to an under-frequency
event is shown in Figure 4(a). As seen, the simulated response is close to the
measured response when the droop is set to 5% (the green curve), which is
consistent with the known control system settings in the SF model. Figure 4(a) also
shows the response of the SF model when the droop is set to 3% (the red trace) and
1.7% (the black trace). As seen, at lower % droop value the FFR response of the SF is
increased, while the time to provide the response is almost unchanged. This,
however, requires that suf cient headroom is available to provide raise FFR. Figure
4(a) shows that the delay observed in the measured response of the SF (the orange
curve) is slightly larger than the simulated response (the green curve). This could be
due to the minor model discrepancies, where some delays present in the actual
plant’s response has not been adequately captured in the SF model.
The active power response of the EMT model of the SF to the over-frequency event is
shown in Figure 4(b). Again, the simulated response closely matches the measured
response when the droop is set to 5% (green trace), consistent with the known
control system settings in the SF model. Also, the gure shows that changing the
droop setting of the model to 3% and 1.7% signi cantly increases the FFR capability
of the SF.

Figure 4 - Simulated and measured responses of SF to (a) an under-frequency event, and (b) an over-
frequency event
4. Comparison of FFR provided by SF and BESS

In this section, the FFR responses by both the SF and the BESS to an under-
frequency event are compared. For this purpose, the measured frequency signal
during the event is fed into the EMT models of both SF and BESS using the playback
setup of Figure 3. The droop and dead-band settings of both SF and BESS are set at
1.7% and ±20 mHz, respectively. Figure 5 shows the overlaid responses of SF and
BESS to the frequency change. As seen, both models respond quite similarly to the
event, when the same settings are used for them. This suggest that SFs can provide
FFR similar to BESSs during under-frequency events, if they have headroom
available via pre-curtailment.
Figure 5 - Overlay of simulated responses of SF and BESS to an under-frequency event
5. Role of FFR in power system security

This section demonstrates the role of FFR in improving overall power system
security. For the purpose of this analysis, FFR from SFs and BESSs is considered. It
should be noted that other technologies such as super-capacitors and fast load
shedding can also provide similar fast response to a change in the frequency.
5.1. Power system model

The wide area EMT model of a gigawatt-scale region of the NEM is used to
demonstrate impacts of FFR on overall power system security. The model includes
detailed modelling of all SGs, IBR, DC interconnector, and all transmission elements.
The model has been benchmarked against measured responses of the actual
network during several events [12]. It is worth mentioning that the analysis carried
out in this paper could have been conducted using RMS-type simulation tools as well,
given that the focus of the paper is frequency stability, however using site-speci c
and benchmarked EMT-type models of individual plants would help accurately
model the temporary energy de cit of the WFs and SFs during the fault recovery
period, which can considerably impact the frequency stability of the grid. Some
short-term transients during and after the fault may not be captured by RMS-type
simulations which could impact the accuracy of the results.
5.2. System conditions

To understand the contribution of FFR to power system security, ve scenarios have
been considered (shown in Table 1). In all scenarios, a fault and disconnection of a SG
coincident with some amount of DPV generation loss has been considered. Droop
response of the SF and BESS are calculated from the point where frequency leaves
the speci ed dead-band. In all the cases, it is assumed that both BESS and SF have
suf cient head-room available. Also, for simplicity and to show effectiveness of FFR
from SF no frequency response is considered from other SGs and WFs in the network.
Table 1 - Simulated case studies on the wide area EMT model of a NEM sub-network
The results of the rst three cases are shown in Figure 6. The key ndings are:
For a contingency size of 190 MW, when only the BESS provides FFR, the
frequency nadir is 49.25 Hz compared to 49.48 Hz when FFR is also available
from SF. The availability of FFR from an SF is dependent on headroom availability
via pre-curtailment.
Other things being equal, availability of FFR from an SF allows an additional 60
MW contingency size, while still maintaining the frequency nadir above the
frequency operating standards. This clearly demonstrates a bene t of the SF’s
FFR for frequency stability of the future grid.
It is observed that in case 3 (the red curves), SF has hit its maximum active
power limit, while BESS has not, while the same droop setting of 1.7% has been

used for both devices. This is mainly due to the fact that the initial generation
level of the SF is 25 MW (which limits its FFR headroom), compared to the 0 MW
in the BESS. Energy availability is required for BESS to provide and maintain FFR
for longer durations.
Figure 6 - Wide area results: (a) frequency (b) FFR from SFs (c) FFR from BESS
The results of cases 2, 4 and 5 are shown in Figure 7. The key ndings are:
Using a wider dead-band of ±150 mHz, the FFR responses of both the SF and
BESS start with around 300 ms delay, compared to when a ±20 mHz dead-band is
used. This leads to a 0.1 Hz lower frequency nadir when a larger dead-band is
applied, for the same contingency size.
When a droop of 3% is used, the frequency nadir is 0.15 Hz lower compared to
when a 1.7% droop is used, keeping all other conditions unchanged.
Figure 7 - Wide area results: (a) frequency (b) FFR from SFs (c) FFR from BESS
5.3. Potential limiting factors of FFR from SF and BESS

FFR from SFs and BESSs rely on reliable and accurate measurement of the
frequency. The time required to measure frequency and act on it can depend on the
frequency measurement algorithm, and measured data update rates. These factors
may vary between manufacturers.
If the frequency is measured and responded to directly at the individual inverter

level of a BESS or SF, the FFR from the SF or BESS may be quicker, compared to a
design where frequency response is managed centrally through the Power Park
Controller (PPC), with a power change command transmitted to each inverter. Delays
associated with communication between the PPC and inverters can add up to 200 ms

or more to response times. Moreover, if the frequency is measured at PPC level, the
speed of the response would depend on the update rate of quantities being
measured and calculated in this central controller.
In the NEM, FFR from BESS has been implemented both at the individual inverter
level and at the central PPC level, depending on the site. In SFs, frequency response
has been implemented at the PPC level only to date.
6. International Experience using FFR

FFR services are used by many other international power system operators. Each
operator has a different way of specifying FFR, suitable to its grid conditions, with
some including renewable sources of generation in the FFR provision framework.
This section provides a summary of FFR utilization in some power systems around
the world.
6.1. ERCOT (Texas)
ERCOT (in Texas) has FFR services implemented from 1 March 2020. Under this
service FFR must be delivered within 0.25 seconds once the frequency is below
59.85 Hz (i.e. 0.15 Hz below nominal frequency of 60 Hz) [13]. Figure 8 shows an
expected response from FFR providers. Earlier, ERCOT already had a type of fast
frequency response product provided by load resources with under-frequency relays
set to trip within 0.5 seconds of frequency reaching or falling below 59.7 Hz. This
service still remains, even after implementation of FFR. At this stage, all of the new
FFR service is provided by storage resources. Resources providing FFR shall be also
capable of governor-like response with ±17 mHz dead-band, and droop setting of
5.0%. Once FFR is deployed, it should sustain for up to 15 minutes or until frequency
recovers to 59.98 Hz (whichever happens rst). Once FFR is recalled, it should
restore within 15 minutes.
Figure 8 - ERCOT FFR [13]

6.2. EirGrid (Ireland)
EirGrid uses ‘Fast Acting Operating Reserve’ to describe FFR. This ‘Fast Acting
Operating Reserve’ is the additional increase in MW output from a generator or
reduction in demand following a frequency event that is available within 2 seconds of
the start of the event and is sustained for at least 8 seconds.
As the majority of FFR resources is provided by wind generators, EirGrid also limits
energy extracted from wind generators. The extra energy provided in the 2-10
second timeframe by the increase in MW output must be greater than the loss of
energy in the 10-20 second timeframe due to a reduction in output power. As shown
in Figure 9, in order to be eligible for FFR, the amount indicated by the blue hatched
area (Power provided) shall be greater than the green hatched area (Power drawn)
[14].
Figure 9 - EirGrid fast acting operating reserve [14]
6.3. National Grid (UK)

National Grid uses ‘Firm Frequency Response’ to describe FFR [15]. FFR represents
the ability to modify generation or demand to compensate for changes in system
frequency within 2-30 seconds, depending on FFR service type. The FFR service is
split into two physical products, Non-Dynamic (static) and Dynamic frequency
response. Static frequency response is activated at a de ned frequency deviation
which is speci ed in agreement with each FFR provider, and no response is required
within the speci ed band. Static response should be provided within 30 seconds and
last for 30 minutes, and hence is categorized as secondary frequency response.
Dynamic frequency response, on the other hand, is continuously provided to manage
second-by-second frequency variations. Dynamic response is automatically
delivered when frequency violates the dead-band of 50Hz ±0.015Hz. It has three
types including primary, secondary, and high frequency responses. Primary dynamic

response should be delivered within 2 to 10 seconds and last for 20 seconds.
Secondary dynamic response should be delivered within 30 seconds and last for 30
minutes. High dynamic response should be delivered within 10 seconds and last
inde nitely, unless agreed otherwise with the provider.
7. Conclusion
This paper provides real-life evidence that SFs have capability to provide FFR. The
response is con gurable and can closely align with that of BESSs. Responses from an
SF during frequency events have been analysed. The real-time observations show
that grid-scale SFs have the potential to provide FFR to the grid, contingent on the
availability of active power headroom when responding to under-frequency. Using
the benchmarked model of a SF, it was shown that SFs can provide similar FFR to
that from a BESS, if they are properly tuned for this purpose and subject to resource
(i.e. headroom) availability. A detailed wide area EMT model of a NEM region was
used to evaluate the impact of FFR from SFs on overall frequency control of the
power system. The results verify that the frequency response of the grid can be
signi cantly improved when the FFR from SFs is available. It was also shown that the
improved frequency stability allows for larger contingency sizes. Besides, it was
shown that the FFR control settings, such as droop and frequency dead-band, are
crucial for determining the effectiveness of the response, where a tighter dead-band
and a smaller droop noticeably improve the frequency stability of the grid.
References
1. IEEE, “Renewables in Australia”, IEEE Power & Energy Magazine, Vol 19, n. 5,
September/ October 2021
2. AEMO, “Transfer Limit Advice, System Strength in SA and Victoria”, Australian
Energy Market Operator, July 2021
3. NERC, "Fast frequency response concepts and bulk power system reliability
needs." NERC Inverter-Based Resource Performance Task Force (IRPTF), March
2020.
4. AEMO, "Fast frequency response in the NEM", Australian Energy Market Operator,
2017.
5. AEMO, “South Australian Electricity Report”, Australian Energy Market Operator,
November 2020.
6. D. Stenclik, M. Richwine, N. Miller, L. Hong, “The Role of Fast Frequency Response
in Low Inertia Power Systems”, CIGRE PARIS 2018.
7. Brogan, P., et al. "Fast frequency response requirements for replacement of
observed generator response during under frequency transients." CIGRE
Symposium, Dublin, 2017.
8. Eriksson, Robert, Niklas Modig, and Katherine Elkington. "Synthetic inertia
versus fast frequency response: a de nition." IET Renewable Power
Generation 12.5 (2018): 507-514.
9. Varma, Rajiv K., and Mohammad Akbari. "Simultaneous fast frequency control
and power oscillation damping by utilizing PV solar system as PV-
STATCOM." IEEE Transactions on Sustainable Energy 11.1 (2019): 415-425.
10. C. Loutan, P. Klauer, S. Chowdhury, S. Hall. “Demonstration of Essential Reliability
Services by a 300-MW Solar Photovoltaic Power Plant”, NREL/TP-5D00-67799.
National Renewable Energy Lab. (NREL), Golden, CO (United States), 2017.
11. V. Gevorgian, B. O’Neill, “Demonstration of Active Power Controls by Utility-Scale
PV Power Plant in an Island Grid”, NREL/CP-5D00-67255. National Renewable
Energy Lab. (NREL), Golden, CO (United States), 2017.

12. Babak Badrzadeh, Z. Emin, Emil Hillberg, David Jacobson, Lukasz H. Kocewiak, G.
Lietz, Marta val Escudero, “The Need for Enhanced Power System Modelling
Techniques and Simulation Tools”, CIGRE Science & Engineering Journal,
February 2020.
13. Julia Matevosyan, “Overview of Renewables in the ERCOT System”, ERCOT, Wind
Integration Workshop Stockholm, October 2018.
14. EirGrid, SONI, "DS3 system services: portfolio capability analysis", EirGrid and
SONI, 2014.
15. National Grid, “Firm Frequency Response (FFR)”, National Grid, December 2017.
Contact Author: N. MODI 


Cse 24

Uploaded by

Document Information

Original Title

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

Cse 24

Uploaded by

Copyright:

Available Formats

CSE N°24

©2022 - CIGRE CSE N°24 February 2022 1

Editorial by Rob Stephen

Mitigation measures for increasing penetration of solar Photovoltaic (PV) systems in

Virtual Synchronous Generator Versus Synchronous Condensers: An

Low-cost Ultrafast Modular HVDC Circuit Breaker

Line Resonance Analysis Implemented in a Cloud-based Monitoring System for

Comparison of electrical clearance between Japan and other countries

Eco-design as a systemic innovation method for offshore substations

Information Synergy: Integrating data across disciplines and functions

Distinctive Downsizing of Cone-Type Insulating Spacer for 245 kV Class GIS by

Synthetic Data – A Solution to Train Diagnostic Systems for High-Voltage Equipment

Limits of electric eld for composite insulators: state-of-the-art and recent

Fast Frequency Response from Transmission-Connected Solar Farms: Australian

©2022 - CIGRE CSE N°24 February 2022 2

Editorial by Rob Stephen

1. Free publication – no nancial obligation to authors;

In addition, we are pleased to publish ve papers submitted by authors that have

©2022 - CIGRE CSE N°24 February 2022 3

All information on how to go about publishing can be found on this page.

There is no charge to publish, the magazine is free download providing the

©2022 - CIGRE CSE N°24 February 2022 4

Mitigation measures for increasing

A. PANDARUM - Eskom Holdings SOC Ltd South Africa

PV penetration - Voltage rise - Network asset congestion - Mitigation measures

©2022 - CIGRE CSE N°24 February 2022 5

The primary challenges of increasing penetration is generally related to the impact

[1] details the results for potential technical impacts of increasing penetration of PV

©2022 - CIGRE CSE N°24 February 2022 6

Figure 1 - Mitigation measures with associated locations

©2022 - CIGRE CSE N°24 February 2022 7

2.1. Smart inverters

©2022 - CIGRE CSE N°24 February 2022 8

The mitigation measure recommended here makes use of function i) mentioned

Figure 2 - Example of tailored Volt-Watt and Volt-Var inverter settings [5]

2.2. Installation of Behind-the-meter Battery Energy

2.3. Adjustment of Off-load Tap Changers in existing

©2022 - CIGRE CSE N°24 February 2022 9

2.4. Installation of Low Voltage Regulator (LVR)

2.5. Network Augmentation (MV & LV) – Transformer

©2022 - CIGRE CSE N°24 February 2022 10

2.6. Adjustment of Zone Substation OLTC

2.7. Installation of MV/LV On Load Tap Changer (OLTC)

©2022 - CIGRE CSE N°24 February 2022 11

Figure 5 - Voltage control logic of OLTCs in transformers [8]

2.8. Combination solutions [4]

©2022 - CIGRE CSE N°24 February 2022 12

©2022 - CIGRE CSE N°24 February 2022 13

©2022 - CIGRE CSE N°24 February 2022 14

Contact Author: A. PANDARUM 

©2022 - CIGRE CSE N°24 February 2022 15

Virtual Synchronous Generator Versus

S. HADAVI, S. PHU ME, B. BAHRANI - Department of Electrical and Computer

©2022 - CIGRE CSE N°24 February 2022 16

Historically, most existing power system networks consist of synchronous generators

Nowadays, there are countless nancial and environmental incentives to increase

In addition to inertia, synchronous generators can provide fault current contribution

From generators prospect, establishing and maintaining a stable connection to a

©2022 - CIGRE CSE N°24 February 2022 17

The rest of this paper is organized as follows. In Section 2, an explanation of current

2. Weak grid stability issues and grid code

Furthermore, IBRs, typically operate in a grid-following mode. In the grid-following

©2022 - CIGRE CSE N°24 February 2022 18