Arc Fault Detection in DC Microgrids

Christian Strobl
E-T-A Elektrotechnische Apparate GmbH
Altdorf, Germany
Email: [email protected]

Abstract—Compared to conventional AC grids, DC electronic protection is reasonable in numerous applications

microgrids demand different switchgear and safety concepts. with voltages greater than 20V in order to break or to limit
Electric arcs across mechanical contacts of circuit breakers overcurrent.
during switching operations and also arc faults within the
installation are more stable in the case of continuous current. Especially if galvanic isolation has to be guaranteed in open
Model based methods are important tools not only for selecting circuit, the use of hybrid devices as a combination of
the appropriate method of converter control, but also for mechanical switchgear and power electronics is a state-of-the-
analyzing the boundary conditions for faults and to develop art-technology to prevent contacts from erosive burning. If a
reliable arc fault detection devices. special circuit design is used, the voltage drop of a switching
arc in a parallel switching chamber can be used as a voltage
Keywords—arc fault detection; circuit breakers; switchgear; supply for the solid-state electronics. In this case no separate
model based methods; large-signal analysis; small-signal analysis feed-in is needed; the electronics are energized only during
switching operation and bypassed in on-state (Fig. 2).
I. INTRODUCTION A main safety issue at higher system voltages is the
With an increasing number of diverse DC microgrid possible danger of fire and personal injury caused by arc faults,
installations the research on specific safety aspects is gaining which are not within specific switchgear, but may be anywhere
more and more importance. Besides the implementation of in the installation at loose contacts or at damaged cables. Arc
qualification processes of components and installation, also a faults are more dangerous and more difficult to detect in the
detailed survey of safety and stability of control methods has to case of direct current compared to AC systems with repeated
be conducted not only for large scale, but also for small scale zero-crossings of current.
microgrids. Especially due to the large number of worldwide
photovoltaic installations and the increasing use of battery Due to the use of storage systems bidirectional flow has to
systems within the last years several investigations have been be considered in safety concepts, specific breaking strategies
carried out about the reasons of reported fire incidents [1] and have to be established for faults within distributed sources.
initiated research programs and standard specifications [2]. Central and distributed control topologies and methods for DC-
DC converters linked to DC buses and for inverters connecting
Arc plasma (Fig. 1), which occurs in mechanical switches to AC grids have to be tested for stability not only in the range
during breaking operations or at faults, is more stable in DC of typical operating points. Also possible cases of severe faults
than in conventional AC installations. Hence DC grids demand have to be included in stability analysis in order to avoid more
extended safety concepts: Mechanical circuit breakers have to serious secondary damage, e.g. due to overcurrent or to
be certified for longer breaking operation by the use of specific overvoltage.
switching chambers or arc quenching methods. Modern
II. CHARACTERISTICS AND MODELS OF ELECTRIC ARCS
Arc plasma between contacts of different voltage in an
electric circuit consists of quasineutral dissociated gas,
originated from air and contact material, and with temperatures
of several thousand Kelvin. The fundamental electrical
characteristics of arcs can be specified as follows [3]:
• A stable electrical arc requires a minimum voltage of
11–14 V across the contact points depending on
material (sum of cathode and anode fall).
• A minimum current of approx. 0.5 A is needed.
• An arc ignition across initially currentless open contacts
needs a higher flash voltage in the beginning.
• An elongation of the column of a stable arc leads to an
Fig. 1. Electrical arc within DC test-setup with photovoltaic source, increase of voltage and to a decrease of current.
operating point of source at approx. 400V, 25A.

978-1-4799-9880-7/15/$31.00 ©2015 IEEE 181

 • High-frequency variations in conductivity caused by III. BASIC PRINCPLES OF ARC FAULT DETECTION
surface abrasion and impact of non-dissociated melted Parallel arc faults across transmission lines or within loads
material (especially in the case of short arc column) and often provoke a rapid increase in current and activate
due to collisions of dissociated particles can be overcurrent circuit breakers. Very fast optical and acoustical
observed. parallel arc fault detection methods within a confined volume
For modeling the quasi-stationary behavior of a stable arc of an electrical cabinet are also well established.
the relation between arc voltage Uarc, the current Iarc and the However, if widespread grid installations have to be
length of the arc column larc can be described with the Ayrton monitored for low-energy series arcs or for parallel arcs within
formula [4], [5]: distributed sources, the examination of measured electrical
variables with respect to the characteristics of the system is
Uarc = a + b larc + (c + d larc) / Iarc (1) essential. In addition to fault plausibility tests the basic
principles of online fault detection methods by current and/or
The parameters a, b, c and d are material- or form- voltage monitoring with appropriate sensor devices can be
dependent; a is minimum arc voltage, b the field strength derived from the fundamental arc properties.
between the contacts, and the sum c + d larc is the current- Depending on the application, in many cases slow
independent part of the arc power. monitoring of electrical parameters for a search of steps in
For the purpose of arc fault detection only a basic voltage or current is sufficient. If voltage sources or converters
plausibility check can be stated with respect to the Ayrton with a fast and stable control of the output-voltage are used and
equation due to unknown arc length and also a large variation the system voltage is not too high, voltage sensing at the loads
range of b, c and d: The existence of a stable arc is impossible, searching for drops of at least 15 V will be a powerful method
if the voltage at a fault is lower than Uarc,min • a, or the current is to find series arcs in the lines of distribution grids.
lower than Iarc,min • 0.5A. If, on the other hand, overload situations with current
The dynamical behavior of an arc can be estimated with the limiting mode or possible low-energy parallel faults have to be
Cassie and Mayr equations, measured time-constants for arcs included in arc fault monitoring within a distribution grid, load
in air vary in a great range between less than 1 µs up to approx. side methods searching for voltage or current drops might be
120 µs [4], [6]. Similar to the quasi stationary approach, a more not sufficient. Additionally source side measurements at the
or less precise modeling is only possible if the boundary central source or converter feeding the distribution grid are
conditions are well known. necessary.
The task will be more complicated, if grids with numerous
fluctuating distributed sources with a great range of possible
operating points and feeding a central converter have to be
monitored. Voltage and current steps within the grid often
cannot be measured directly at the inverter-input. Inherent
reactions of the sources or the control systems and damping
effects may provoke only very small steps at a central sensor
location depending on the operating point of the system (Fig.
3). On the other hand, in many cases it is not possible due to
technical or financial reasons to use a huge number of separate
sensors looking for rapid changes in voltage or current at each
source module.
For these applications measuring and analyzing of high-
(a)
frequency effects at one or at least at a few sensor locations
often is the method to choose or to add: The arc-typical noise
effects can be interpreted as a multiband source at the fault
location. The resulting high-frequency signals are distributed
within the grid with respect to the damping characteristics of
the components and lines.
Usually analog inductive or capacitive coupled sensor
front-ends are used to measure specific high-frequency bands.
The data thus received is preprocessed and digitized. Digital
pattern recognition methods have to distinguish between arc-
(b) typical random noise and other more regular high frequency
effects on the line, e.g. AC-ripple caused by converter
Fig. 2. (a) Hybrid DC-switch PVSEC for applications up to 1000V, 30A, switching frequencies and their harmonics. If a noise event is
with additional arc fault detection sensor (typically used as firefighter switch categorized as a presumable arc fault, a shut-off or a suitable
for photovoltaics with the option of remote control). (b) Hybrid switching
principle with a solid-state switch in parallel to one of three mechanical selective switching or control operation is initiated.
switching chambers for all-pole disconnect.

182
 If – as a fundamental condition for the following – the
dynamics of the analyzed DC grid can be described with a
separate modeling of slower large-signal and faster small-
signal effects, a similar approach for fault models to be
included is favorable:

A. Large-Signal Analysis: Testing for Arc Fault Boundary

Conditions and for System Response
Modeling large-signal dynamics of a burning arc resulting
from thermodynamic processes and contact erosion is really
challenging even for a well-known contact situation. It is quite
impossible to do this for a fault location with unknown
parameters. However, after a short time-span of ignition with a
rapid change in voltage across the contacts, the corresponding
time constants for large-signal dynamics of steady burning arcs
are relatively high. For that reason, especially arcs of short
length can be modeled as switched voltage drops for short-time
and large-signal model purposes with sufficient accuracy.
Hence it is reasonable for large-signal analysis to calculate the
system reaction to a certain number of different voltage steps
according to a variation of arc lengths at possible fault
locations (Fig. 5).
Further simplifying, the large-signal system characteristics
can be modeled with a short-time quasi-stationary approach, if
the following conditions are fulfilled:
• The system behavior can be characterized with different
operation modes.
Fig. 3. Series arc fault in a photovoltaic system (0.8 s – 3.5 s, manually
opened and closed contacts): no steps in current and voltage at inverter input • The transition between different operation modes is
observable, high frequency sensing is required (3rd subplot). sufficiently fast (e.g. tripping of breakers, switching).
• The large-signal dynamics of the system are quite slow
IV. INCLUDING FAULT MODELS INTO GRID MODELS within each operation mode.
The electrical characteristics of arcs can be used for fault • The system voltage is relatively high compared to
plausibility tests and for the basic principles of fault detection assumed arc voltages within a minimum time-span,
based on current and/or voltage monitoring at appropriate which is needed for detection (At slowly opening
locations within the grid: contacts, the series arc voltage usually has its minimum
Especially in DC microgrids with a large variety of possible of approx. 15-20 V shortly after ignition, when the
sources and loads with a great number of operating points it contact gap is still quite small.).
makes sense to use model-based methods in order to analyze With the comparisons of the calculated resulting fault
the topology- and operating-point-dependent vulnerability to current with the minimum value for a stable arc, and of the
faults and to find a suitable arc fault detection method. resulting power with critical values for the ignition of
With respect to the characteristics of sources, loads and insulation material, an arc fault plausibility check and a
DC-DC-converters linking different voltage levels, model calculation of the fire risk can be performed for different
based methods can present crucial information about the possible fault locations and system modes.
boundary conditions for stable arcs and for the propagation of
arc-specific wideband-noise within the grid (Fig. 4). B. Small-Signal Analysis:Calculating the Noise Propagation
within the Network
In the first instance they are helpful for system design and
Small-signal high-frequency arc characteristics are also
for control methods in order to avoid or to de-energize harmful
quite complicated to model; pink colored or Brownian
arcs in an inherent way. They provide essential information to
broadband-noise is assumed [8]; an equivalent circuit with a
develop reliable arc fault detection devices, avoiding numerous
multi-frequency voltage source and a resistor can be used. The
experiments at different operating points and giving hints for a
typical amplitudes have to be estimated by experiments for
precise feature selection in pattern recognition methods in time
different contact situations.
and frequency domain. Furthermore, switching concepts or
control strategies in the case of faults referring to the specific Small signal models of the grid and its components can be
system characteristics can be established. built up in a modular way using the help of circuit simulation
software and experimental methods like impedance

183
spectroscopy. The damping characteristics of the transmission Reliable arc fault detection devices have to be sensitive for
lines from the assumed fault locations and other high- arc faults avoiding false tripping due to other high frequency
frequency sources (switched power-electronics) to the sensor sources. The linearized small-signal model gives important
devices can be calculated. Using the superposition principle the information for selecting the best frequency-band to analyze
resulting amplitudes can be estimated and compared for and hints for appropriate analog preprocessing and pattern
different frequencies and a great number of operating points. recognition methods.

Fig. 4. Including fault models into large-signal and small-signal system models: calculation of fault boundary conditions, system response and noise propagation
for fault analysis and arc fault detection methods. Modeling of dynamic large-signal and noise effects of arcs is difficult.

Fig. 5. Large-signal model (quasi-stationary, if possible) and linearized small-signal model for arc fault detection.

184
 V. EXAMPLE: MODEL-BASED ARC FAULT DETECTION IN • faults in parallel to a cell-string antiparallel to the
PHOTOVOLTAIC APPLICATIONS bypass diode (needs a damage within the junction box,
Photovoltaic systems are typical and widely-used examples arcs are not possible for most multi- or polycrystalline
for multi-source DC-systems with fluctuating power. They modules due to the too low cell-string-voltage),
have a great variety of generator and converter topologies with • faults in parallel to a cell-string in series to the bypass
different control strategies and are influenced by a large diode at negative cell-string voltage (loose contact of
number of input parameters, like solar irradiance, partial the diode and partial shading required for arcs).
shading, temperature and grid or DC-link voltage. Due to the
very huge number of possible structures and operating points, Modeling different system topologies and analyzing the
model based methods are an essential tool for risk estimations noise propagation give hints for the sensitivity design for arc
and for the development of arc fault detection devices [9], [10]. fault detection devices with respect to different converter
topologies and control methods and for a field of
The correlation of cell current and voltage of solar cells can configurations as diversely as possible.
be described with one- or two-diode-models. The dependencies
to irradiation and temperature are estimated using material-
specific data and data sheet information, sub-models for diode-
breakdown and temperature-rise are includable.
The resulting characteristic curves for cell-specific sets of
input-parameters can be calculated iteratively and added with
the help of interpolation methods successively to curves of cell-
strings including bypass-diodes, of modules and of strings.
Usually a single string or several strings in parallel are
connected to a maximum-power-point-tracker of a DC-DC- or
a DC-AC-converter (inverter).
For different locations large-signal-models of possible
faults can be included as voltage drops with variable
amplitude (Fig. 6). By considering also lower voltage drops as
needed for a stable arc, glowing effects can be analyzed
additionally.
The transition of the quasi-stationary working point of the
converter input has to be calculated. A simple load model of
the converter based on a constant transmission-ratio for some
hundred milliseconds is convenient in most cases due to the
usually very slow maximum-power-point-tracking, as
experiments showed (Fig. 7). The effects on the line currents
and on the specific cell voltages can be calculated.
Furthermore an estimation of inherent system reactions is
possible.
For an estimation of the small-signal damping of the
transmission lines, the modules and the converters, offline
impedance spectroscopy gives basic information (see also Fig. 6. Characteristic curves of photovoltaic string with fault in series,
[11], [12], [13]). As it can be seen in Fig. 8, the small-signal different fault voltages.
impedance of the modules shows a strong dependency
especially to the panel voltage, which usually changes, when
an arc occurs.
Using a large variety of system topologies for the model,
basic information for arc faults and arc fault detection methods
can be gained, limiting the number of required experiments.
Series and parallel faults in the cabling, and also faults within
modules or junction boxes due to bad soldering-points or faulty
bypass-diodes have been considered:
• series faults in a string or in a collecting line (most
typical fault, e.g. caused by a loose plug connection),
• faults in parallel to one or several modules of a string
(needs a double damage of insulation in ungrounded
systems),
Fig. 7. Models for transition of operation points when fault occurs.

185
 • Alternatively methods of distributed sensing with
devices measuring the voltage at each module and with
a connection to a central control unit with a separate or
a power-line communication bus can be used, but
usually are more expensive.

VI. OUTLOOK
Further research in this direction is carried out in
collaboration with Friedrich-Alexander-Universität Erlangen-
Nürnberg, Germany, and with financial support by the
Bavarian Research Foundation (Bayerische Forschungs-
stiftung) within the next years for different DC system
voltages.
Online system identification and machine learning methods
will be developed to be implemented in electronic circuit
breakers to gain more precise information of the monitored
system and its range of working points. With an integrated
system model it should be possible to identify different kinds
of faults, like shortcuts, gradual malfunction and also arcs in an
earlier stage compared to conventional protection devices; a
better prevention against further damage of the grid and its
components can be gained.

REFERENCES
[1] TÜV Rheinland and Fraunhofer ISE, www.pv-brandsicherheit.de
(German research program about fire risks in photovoltaic installations).
[2] UL 1699B, Outline of Investigation for Photovoltaic (PV) DC Arc-Fault
Circuit Protection, Issue Number 2, January 14, 2013.
[3] P. Slade (Ed.), Electrical Contacts - Principles and Applications. New
York, Basel: Marcel Dekker, Inc., 1999.
[4] A. Erk and M. Schmelzle, Grundlagen der Schaltgerätetechnik. Berlin,
Heidelberg, New York: Springer Verlag, 1974.
Fig. 8. Small-signal impedance of a photovoltaic module with cabling as a
[5] C. Strobl and P. Meckler, “Basic Experiments for Detecting Arc Faults
function of DC-voltage and solar irradiance and at different temperatures.
in DC Aircraft Networks,” Proc. 24th ICEC, St. Malo, 2008, pp. 353-
358.
Additionally the following results can be stated for [6] M. Lindmayer, Schaltgeräte Grundlagen, Aufbau, Wirkungsweise.
photovoltaic systems as typical examples for DC microgrids Berlin, Heidelberg, New York: Springer Verlag, 1987.
with distributed sources: [7] S. Bacha, I. Munteanu and A.I. Bratcu, Power Electronic Converters
Modeling and Control. London: Springer, 2014.
• Sensor modules searching for high-frequency arc [8] S. Bieniek, Impedanzspektroskopie an Photovoltaikanlagen. Berlin:
characteristics are essential for central methods at the Mensch und Buch Verlag, 2012.
inverter or combiner box. [9] C. Strobl, “Störlichtbögen in Photovoltaikanlagen - modellbasierte
Szenarien- und Detektierbarkeitsanalysen,“ 22. Albert-Keil-Kontakt-
• For large photovoltaic systems with a lot of strings in seminar, Karlsruhe, 2013, pp. 64-73.
parallel, high-frequency sensors for each string or for [10] C. Strobl, “Arc Fault Detection - a Model-based Approach,” Proc. 27th
small groups of strings are needed. A single sensor at ICEC, Dresden, 2014, pp. 367-373.
the collecting line may not gain sufficient performance [11] J. Johnson, D. Schoenwald, S. Kuszmaul, J. Strauch and W. Bower,
due to the low quotient of arc noise amplitudes and “Creating Dynamic Equivalent PV Circuit Models with Impedance
inverter ripple at the sensor location. Spectroscopy for Arc Fault Modeling,” 37th IEEE PVSC, Seattle, 2011,
pp. 2328-2333.
• Categorizing detected arcs in series and parallel ones or [12] J. Johnson, J. Strauch, S. Kuszmaul, W. Bower and D. Schoenwald,
finding the arc location is very difficult for central “Characterizing PV Arcing Conditions with Impedance Spectroscopy
and Frequency Response Analysis,” 26th EU PVSEC, Hamburg, 2011,
sensor devices in the string or in the string collector, pp. 3751-3756.
especially in the case of partial shading and the [13] D. Schwander, “Dynamic Solar Cell Measurement Techniques: New
activation of bypass diodes. Small Signal Measurement Techniques,” Space Power, Proceedings of
the Sixth European Conference, Porto, 2002, pp. 603-608.

186