290 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO.

3, SEPTEMBER 2000

Determining the Relative Effectiveness of Islanding

Detection Methods Using Phase Criteria and
Nondetection Zones
Michael E. Ropp, Member, IEEE, Miroslav Begovic, Senior Member, IEEE, Ajeet Rohatgi, Fellow, IEEE,
Gregory A. Kern, R. H. Bonn, Sr., and S. Gonzalez

Abstract—Islanding of a utility-interactive photovoltaic (UIPV)

system occurs if the UIPV system continues to power a section of
the utility system after that section has been disconnected from
the utility source. Since islanding creates hazards for personnel
and equipment, UIPV systems are required to detect and prevent
it. It is desirable to have a simplified method of determining which
islanding detection methods (IDM’s) are most effective. In this
paper, a previously described method for finding the nondetection
zones (NDZ’s) of IDM’s is experimentally verified. This method is
used to determine the NDZ’s of several common IDM’s. These re-
sults indicate that, of the IDM’s discussed in this paper: 1) Sandia
Frequency Shift (SFS) is most effective; 2) the worst-case loads
are low-power loads that are near resonance at the line frequency
and have a large capacitance and small inductance (a high value Fig. 1. System configuration for explaining islanding.
of the quality factor ).
SFS Sandia Frequency Shift
LIST OF SYMBOLS AND ABBREVIATIONS SMS Slide-Mode Frequency Shift
AFD Active Frequency Drift (frequency bias) UFR Underfrequency relay
AFDPF AFD with positive feedback (aka SFS) UIPV Utility Interactive Photovoltaic (system)
Load capacitance Instantaneous node voltage
Normalized load capacitance Frequency of node voltage (rad/sec)
Chopping fraction (AFD parameter) Utility voltage frequency (rad/sec)
Phase threshold (PJD parameter)
Transfer function used to implement SMS I. INTRODUCTION
IDM Islanding Detection Method
PV system output current
Gain (SFS parameter) I SLANDING of a UIPV system occurs when a section of the
utility system is isolated from the main utility voltage source,
but the UIPV system continues to energize that section. Con-
Load inductance (H)
sider the system drawn in Fig. 1. A UIPV system, consisting of
OFR Overfrequency relay
a PV array and a PC’s (or inverter), is connected to the local
Real power (W)
utility system at Node . The utility voltage source at the right
PCS Power Conditioning System (aka inverter)
can be disconnected from Node by the switch. A local load
PJD Phase Jump Detection
assumed linear) is also connected at Node . Power flows are
PLL Phase Locked Loop
defined in the figure. , the reactive power being supplied
POCC Point Of Common Coupling
by the PV system, is typically zero. Consider a case in which
PV Photovoltaic
, and (a unity power factor
Load reactive power (VAR)
load). Under this condition, there is no current flowing through
Load quality factor
the switch, and and are both zero. Thus, Fig. 1 illus-
Load resistance ( )
trates that if the switch were opened under this condition no
change would occur at Node ; that is, the voltage at Node
Manuscript received May 10, 1999. would be the same before and after the opening of the switch.
M. E. Ropp is with the Electrical Engineering Department, South Dakota This means that there is no change that could be used as a signal
State University, Brookings, SD 57007.
M. Begovic and A. Rohatgi are with the School of Electrical and Computer that islanding is occurring. The PV system continues to operate,
Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0250. Node is held at high potential, and islanding is not prevented.
G. A. Kern is with Ascension Technology Inc., Boulder, CO 80301. Islanding is undesirable for several reasons [1]–[3]. Probably
R. H. Bonn, Sr. and S. Gonzalez are with Sandia National Laboratories, Al-
buquerque, NM 87185. the most important reason is that utility personnel may be un-
Publisher Item Identifier S 0885-8969(00)08436-9. aware that the portion of the utility system including Node
0885–8969/00$10.00 © 2000 IEEE
ROPP et al.: DETERMINING THE RELATIVE EFFECTIVENESS OF ISLANDING DETECTION METHODS 291

Fig. 3. Block diagram representation of PV system and load after the switch
is opened.

of the NDZ’s of certain islanding prevention methods within

Fig. 2. A nondetection zone defined within the “power mismatch space.” this space [6]. This procedure is used here to obtain the NDZ’s
for several common islanding prevention methods, and experi-
mental verification of the results is provided. These results not
is still being energized by equipment on the customer’s side of
only suggest that the Sandia Frequency Shift method is the most
the point of common coupling (POCC). They are therefore un-
effective of those discussed here, but they also indicate that an
knowingly exposed to an electric shock hazard. Another impor-
NDZ in the power mismatch space taken at a single load condi-
tant reason is that if the PV system drifts slightly out of phase
tion can lead to incorrect conclusions about the effectiveness of
with the utility voltage source during islanding, large surge cur-
islanding prevention methods.
rents can flow upon reconnection, and these can damage the PV
system, customer loads, or utility equipment. In spite of the fact
II. PROCEDURE
that the likelihood of islanding is very small, due to the seri-
ousness of these risks, utilities and standards-making bodies re- Several islanding detection techniques that rely on a change
quire that PV PCS’s be equipped with specific islanding detec- in the frequency of the Node a voltage to detect islanding were
tion and prevention schemes such that the PCS disconnects the studied analytically, using simplified, closed-form expressions
PV system from Node when islanding is detected. called phase criteria. The phase criteria were derived and
Over the years, several control schemes for PV PCS’s presented in an earlier paper [6]. However, the principles upon
have been devised to reliably detect islanding. Examples which the derivations were based are as follows. For those
include standard over/undervoltage and over/underfrequency islanding prevention techniques which rely on a frequency
relaying (OVR/UVR and OFR/UFR), phase jump detection change, the islanding-detection behavior of the system is
(PJD) [3], [4], slide-mode frequency shift (SMS) [3], [5], and governed by the behavior of the phase-locked loop (PLL) that
active frequency drift (AFD), or frequency bias [6]. Utilities, synchronizes the PV system’s output current to the utility
standards-making bodies, and PCS manufacturers have a voltage. The PV and load real powers are assumed to be
common interest in determining which of these methods is matched . When islanding begins, the PV system
“best;” that is, which one detects islanding most reliably. Such and load may be modeled as shown in Fig. 3. The Node
a determination may be made through the use of nondetection voltage, , is determined by the Ohm’s Law response of the
zones (NDZ’s). NDZ’s are regions in an appropriately defined load. This voltage is monitored by the PV PCS. The PLL
space in which the islanding detection scheme under test fails inside the PCS will cause the frequency of the output current
to detect islanding. Several definitions of NDZ are possible. of the PV system, , to change until there is no phase shift
In the past, NDZ’s have usually been defined in a “power between and . When and are in phase, no further
mismatch space” like that shown in Fig. 2, in which the axes change in the frequency of occurs; a steady-state frequency
are defined in terms of and as defined in Fig. 1. is reached. However, the phase shift between the PV output
PCS’s were typically tested under full-load (maximum current and the Node voltage is simply the phase of the
power) conditions, and with specific values of , , and . load, which is a known function of frequency. Therefore, the
Frequently, the test load was purely resistive, so that steady-state condition may be described by an expression
and . and were varied from zero up to the level in terms of the parameters of the particular islanding prevention
at which the islanding detection method reliably detected the method and the , and of the load. These expressions are
islanding condition. The region very near in the phase criteria. (A brief description of the derivation of the
which the change in the Node voltage was insufficient for the phase criterion for AFD is given in the Appendix to this paper.)
islanding detection method to operate, as shown in Fig. 2, was The phase criteria for several common islanding prevention
defined as the NDZ. Tests of this type have been used to declare schemes, along with the methods for using them, are given in
that certain islanding prevention methods or combinations of Table I. The methods considered are OFR/UFR, PJD, SMS,
methods have no NDZ at all. However, such declarations on the AFD, and Sandia Frequency Shift (SFS). The details of the is-
basis of this type of testing are based on the assumption that a landing prevention methods are given elsewhere [6], [7]; how-
full-load, – test with a small value of and a large ever, the reader should be aware that all of these methods rely
value of represents a worst-case scenario, an assumption on a change in the frequency of the voltage at Node to detect
which previously had been unproven. islanding. In the Table, is the frequency of the voltage at Node
In this paper, we propose a new NDZ definition in terms of and is the utility voltage frequency. Several design param-
a “load parameter space” (to be defined shortly). We have pre- eters also appear in the Table: is the phase threshold used in
viously described a simple procedure for fording the locations PJD (commonly between 2 and 5 [6]), is the transfer
292 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO. 3, SEPTEMBER 2000

TABLE I
PHASE CRITERIA FOR SEVERAL ISLANDING PREVENTION METHODS
(FROM [6], [7])

Fig. 4. Mapping of the NDZ of the AFD (frequency bias) method of islanding
prevention in a load parameter space, for three different values of load
resistance. The NDZ of the SMS method for R = 14:4
is also shown.

Fig. 5. Mapping of the NDZ of the SMS method of islanding prevention

in a load parameter space. NDZ’s are shown for three different values of
load resistance. In these calculations, the G(j!) used is that of a (formerly)
function that implements the SMS method, is the “chopping
fraction” used in the AFD method, is a gain used in SFS, and commercially-available PCS [3].
is defined as .
The phase criteria can be used to map the NDZ of an is- boundaries of the NDZ of the OFR/UFR alone (dashed hori-
landing prevention method into a load parameter space. For zontal lines) and the NDZ of the SMS method for
each point in the load parameter space, the steady-state fre- (for comparison). The narrowing of the NDZ’s at very small
quency of the PV system may be calculated using the phase values (corresponding to very large values) is due to the
criterion for the particular islanding prevention method under sharpening of the magnitude response of the load in that
study. If the steady-state frequency lies within the trip thresh- region. The magnitude of the load changes so rapidly with fre-
olds of the OFR/UFR, taken throughout this study to be those set quency that, as the frequency changes slightly but the current
by the new proposed IEEE-P929 standard ( Hz [2]), then amplitude remains constant, the voltage changes sufficiently to
that load lies within the NDZ of that islanding prevention cause an undervoltage or overvoltage trip. Note that the NDZ
method. If not, the load lies outside the NDZ. Note that the phase bends upward, into a range of leading-power-factor loads, as
criteria can predict NDZ locations, but they cannot predict the the inductance increases (and decreases). The phase criterion
values of run-on time (the time elapsed between the opening of predicts that AFD could be made to fail for loads (a fact al-
the switch and the detection of islanding by the PCS). ready experimentally verified [8]), but it also predicts that AFD
can be made to fail for unity power factor loads with small
III. ANALYTICAL AND MODELING RESULTS and large .
Using the phase criteria, the NDZ’s of the islanding preven- The NDZ’s of SMS (Fig. 5), PJD (Fig. 6), and SFS (or
tion methods in Table I were mapped into a load parameter space AFDPF, Fig. 7) with a small amount of frequency bias were
defined by the load “normalized capacitance” vs. the similarly mapped. The SMS mapping in Fig. 5 shows behavior
load inductance . is defined as the ratio of the load similar to that of AFD, except that above a certain threshold
capacitance to the capacitance that resonates with the load in- SMS has no NDZ at all. Below that threshold , the NDZ
ductance at the utility line frequency: of SMS is the same as that of the OFR/UFR acting alone, and
the NDZ narrows at very small (large ) values due to the
(1) sharpening of the load’s magnitude response. The NDZ plot of
PJD shows the opposite behavior; PJD’s NDZ becomes wider
Fig. 4 shows a mapping of the NDZ of the AFD method. and wider as increases and decreases. In fact, for practical
The boundaries of the AFD NDZ for three different values of values of PJD appears to be at best only as effective as the
resistance (3.6, 14.4, and 57.6 ) are shown, along with the OFR/UFR acting alone.
ROPP et al.: DETERMINING THE RELATIVE EFFECTIVENESS OF ISLANDING DETECTION METHODS 293

Fig. 8. Experimental setup used to map the NDZ of SFS.

PJD, the worst case loads (that is, those that lead to the longest
run-on times) are predicted to be loads with a relatively low real
power consumption (small ), a small value of , and a large
Fig. 6. Mapping of the NDZ of the PJD method of islanding prevention in a
load parameter space, obtained using phase criteria. In these calculations,  = value of . This may be expressed in terms of the quality factor,
2 [3]. or -factor, of the circuit. (This is not the same as the
reactive power consumption of the load, denoted or .)
The -factor is given by
(2)
and is the ratio of the amount of energy stored in the load’s re-
active elements to the amount of energy dissipated in the load’s
resistance. (For example, for , there is twice as much en-
ergy stored in the and of the load as is being dissipated in
.) Loads that are near resonance at and have a high -factor
are the ones that cause difficulty in islanding detection.
It is clear from the phase criteria and the NDZ mappings that
the steady-state frequency of depends on the load pa-
Fig. 7. Mapping of the NDZ of the SFS method (AFDPF in the figure). For rameters. Unfortunately, the level of real or reactive power mis-
comparison, the NDZ of AFD is also shown, as are the NDZ’s of SMS and the match is not uniquely determined by load parameters. Specifi-
OFR/UFR.
cally, the reactive power consumption of the load is given by

However, the SFS method (Fig. 7) has an extremely narrow (3)

NDZ, encompassing fewer loads than the NDZ of any
Equation (3) clearly shows that there are infinitely many combi-
other method considered here. Thus, the phase criteria predict
nations of and that will yield the same , and will there-
that SFS is the most effective of the islanding prevention
fore map onto the same point in the power mismatch space for a
methods discussed here. Because of the frequency bias, the
given . Some of these combinations result in nondetection of
NDZ bends upward into a range of leading power factor
islanding, while others do not. Therefore, an NDZ in the power
loads as increases. However, this NDZ is so narrow that its
mismatch space measured at a load other than the worst-case
boundaries appear to lie nearly atop one another until smaller
load can lead to misleading conclusions about the effectiveness
values (larger values) are reached. The NDZ is much
of the islanding detection method under test.
wider at small (large ) values, but it is still narrower than
that of any other method. From a practical standpoint, one
might expect that this NDZ shape would suggest that it would IV. EXPERIMENTAL RESULTS AND DISCUSSION
be nearly impossible to obtain a long run-on time for a load An experiment was conducted to verify the accuracy of the
with large and small because of the required level of NDZ predictions made by the phase criteria. The experimental
accuracy of the component values coupled with experimental setup used is shown in Fig. 8. The PCS in this setup used the
variability. The model also predicts a narrowing of the NDZ SFS method. The load inductance was set to a “high” or “low”
for decreasing values. Therefore, experimental run-on times value, and the load capacitance was then swept over a narrow
could also be expected to decrease for higher-power loads. The range near . This procedure was repeated for “high” and
phase criterion thus predicts two behaviors for a PCS equipped “low” load resistance values. The values used in the experiment
with SFS with frequency bias: a) the NDZ should contain only are given in Table II, along with the results in terms of the value
leading power factor loads for “large” ; of at which the longest run-on times were obtained. The
b) long run-on times may be obtained for loads with “small” experimental results are shown graphically in Figs. 9 and 10.
, but a long run-on time would be practically impossible to Fig. 9 shows the run-on time of the SFS-equipped PCS as
obtain for a “large” . Similarly, long run-on times should be a function of for “small” and “large” values (96.6
very difficult to obtain for “small” , but for “large” longer and 345 mH). Both sets of data are for the “high” value
run-on times should be possible. (108.1 for mH, 134.4 for mH). The
In general, for all of the islanding prevention methods that model predicted that the longest run-on times would be ob-
depend on a frequency change for islanding detection except tained for loads with , that the at which
294 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO. 3, SEPTEMBER 2000

TABLE II
PARAMETER VALUES FOR NDZ MAPPING EXPERIMENT

Fig. 11. Results of islanding testing on an AFD-equipped, commercially-

available PCS feeding a purely resistive load.

Fig. 9. Run-on time of the SFS-equipped PCS as a function of C , for

two different values of R.

Fig. 12. Results of islanding testing on the same PCS used in the tests in
Fig. 11, but now with varying L(C = C ).

as a function of and (“high” or “low”). Run-on times in-

crease for decreasing and for increasing . Thus, as pre-
dicted, high- -factor loads are the worst-case loads in terms of
islanding detection. This general trend has been confirmed in
repeated experiments [9]. However, the reader should be aware
that although the trend has been consistently demonstrated, both
longer and shorter maximum run-on times have been experi-
mentally observed.
The shortcoming of the power mismatch space representa-
tion of NDZ’s of islanding prevention methods is now clear.
Fig. 10. Results of NDZ mapping experiment, demonstrating the trend in One could, for example, test an AFD-equipped PCS at full load
run-on times vs. L and R. The islanding detection method is SFS.
(small value of ) and with a purely resistive load ( ,
, -factor = 0), vary and , and conclude that
maximum run-on times were obtained would increase as in- AFD has no nondetection zone. This conclusion is clearly false,
creased, and that maximum run-on times would decrease as as indicated by the AFD NDZ in Fig. 4. Since the NDZ has been
increased. The latter two of these trends are clearly evident in supported by experimental observations, we may conclude that
the data. However, maximum run-on times are actually obtained AFD should always detect islanding for purely resistive loads,
for in the “small” case. This is due to the influence but will fail for high- , resonant RLC loads.
of the large capacitors in the output stage of the PCS, an effect To further verify this claim, a commercially available PCS
which is not accounted for in the phase criteria. The NDZ loca- equipped with AFD was tested under different load conditions.
tions predicted by the phase criteria may thus need to be shifted These experimental results are shown in Figs. 11 and 12. In
slightly toward lower values. However, the location and Fig. 11, the load is purely resistive, and is varied. Some
shape of the NDZ appear to be predicted quite well. Fig. 9 also scatter in run-on times was observed, but in spite of this the
illustrates the effectiveness of the SFS method, as indicated by converter always detects islanding and ceases operation within
the extremely narrow range of values over which high 0.14 seconds (less than 9 line cycles at 60 Hz), far less than the
run-on times are obtained. 2 seconds allowed by IEEE-P929 [2]. This behavior is consis-
Fig. 10 gives further corroboration of the phase criterion pre- tent with that predicted by the phase criterion for AFD (Fig. 4).
dictions. This Figure shows the longest run-on times obtained However, as shown in Fig. 12, the situation is entirely different
ROPP et al.: DETERMINING THE RELATIVE EFFECTIVENESS OF ISLANDING DETECTION METHODS 295

fundamental component of this waveform equal to 0.5 [6],

[7]. If is assumed small, then the amount of distortion of
introduced by the zero times is also small. Thus, the AFD
waveform is reasonably approximated by its phase-shifted fun-
damental component. From the system in Fig. 3 and an under-
standing of the operation of the PCS’s PLL, it may be deduced
that if the PCS were equipped with AFD the frequency of
will increase until the phase shift of the load “un-does” the phase
shift between and its fundamental. Thus,

Fig. 13. i waveform for AFD.

when the load has a unity power factor but is not purely resis-
tive. In these tests, . Notice that, as the -factor which is the phase criterion given in Table II, with
of the RLC load increases (that is, decreases and increases, .
maintaining near resonance at the line frequency), the run-on
time of the PCS also increases. The scatter in run-on times is REFERENCES
primarily due to experimental variability, as discussed previ- [1] A. Kitamura, M. Okamoto, F. Yamamoto, K. Nakaji, J. Matsuda, and K.
ously. The observed results are consistent with the predictions Hotta, “Islanding phenomenon elimination study at Rokko Test Center,”
suggested by the results in Fig. 4. in Proceedings of the 1st IEEE World Conference on Photovoltaic En-
ergy Conversion, 1994, pp. 759–762.
[2] IEEE-P929: Recommended Practice for Utility Interface of Photo-
V. CONCLUSION voltaic (PV) Systems, Draft 9 of IEEE Prestandard P929, December
1998.
This work demonstrates that phase criteria (expressions de- [3] R. A. Jones and T. R. Sims, “Investigation of potential islanding of
rived from the conditions at which the PLL of the PV PCS de- dispersed photovoltaic systems,” Sandia National Laboratories, Albu-
querque, NM, Sandia National Laboratories Report SAND87-7027,
tects a zero phase error between the PV output current and ter- 1988.
minal voltage) are highly useful in determining the relative merit [4] H. Kobayashi, K. Takigawa, and E. Hashimoto, “Method for preventing
of islanding prevention methods that rely on a frequency change mismatch islanding phenomenon on utility grid with a number of small
scale PV systems,” in Proceedings of the 21st IEEE Photovoltaic Spe-
for islanding detection. In spite of their simplicity, the phase cri- cialists Conference, 1991, pp. 695–700.
teria predict the ranges of loads that lead to detection failure with [5] S. Yuyuma, T. Ichinose, K. Kimoto, T. Itami, T. Ambo, C. Okado, K.
good accuracy. In addition, clear evidence is provided that the Nakajima, S. Hojo, H. Shinohara, S. Ioka, and M. Kuniyoshi, “A high-
speed frequency shift method as a protection for islanding phenomena of
SFS method is the most effective islanding prevention method utility interactive PV systems,” Solar Energy Materials and Solar Cells,
among those discussed here. Finally, this work also establishes vol. 35, pp. 77–486, 1994.
clearly that nondetection limited range of loads can give a mis- [6] M. Ropp, M. Begovic, and A. Rohatgi, “Prevention of islanding in grid-
connected photovoltaic systems,” Progress in Photovoltaics, Summer
leading picture of the effectiveness of an islanding prevention 1999, to be published.
method. Mappings should be made in a load parameter space [7] M. Ropp, “Design issues for grid-connected photovoltaic systems,”
such as the space used here, or several power mis- Ph.D. dissertation, Georgia Institute of Technology, Atlanta, GA,
December 1998.
match mappings should be made with different values of RLC. [8] G. Kern, “SunSine300: Utility interactive AC module anti-islanding test
Furthermore, in order to obtain an accurate test of the effec- results,” in Proceedings of the 26th IEEE Photovoltaic Specialists Con-
tiveness of an anti-islanding method that depends on frequency ference, 1997, pp. 1265–1268.
[9] G. Kern and Ascension Technology, Inc., “Status report #6: Sandia anti-
change for islanding detection, it is suggested that the PCS be islanding investigation,”, contract report to Sandia National Laborato-
tested with loads with high values of the -factor. The data pre- ries, May 27, 1998.
sented here indicate that tests should include a sweep of
over the range 0.9 to 1.1, in order to ensure that the NDZ is
accurately located. Michael E. Ropp (S’94–M’99) received the Ph.D. in EE from the Georgia In-
stitute of Technology in December, 1998. He is currently an Assistant Professor
of Electrical Engineering at South Dakota State University. His research inter-
APPENDIX ests include power electronics, renewable energy, and electric vehicles.
CURSORY DESCRIPTION OF THE DERIVATION OF THE PHASE
CRITERION FOR AFD
AFD is implemented by adding short zero or “dead” times Miroslav Begovic (S’87–M’89–SM’92) received the Ph.D. in EE from Virginia
into the PV system output current, as shown in Fig. 13. These Tech in 1989. He is presently an Associate Professor in the School of Electrical
and Computer Engineering at Georgia Institute of Technology. Dr. Begovic’s
dead times have length . The AFD waveform has neither odd research activities are in computer applications in power systems monitoring
nor even symmetry; therefore, there will be a phase shift in the and control, and PV systems.
296 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 15, NO. 3, SEPTEMBER 2000

Ajeet Rohatgi (M’78–SM’86–F’91) received the Ph.D. degree in metallurgy R. H. Bonn Sr. is currently the PV balance of systems Project Leader for the
and material science from Lehigh University, Bethlehem, PA, in 1977. He is a Sandia National Laboratories PV program. He is responsible for developing
Regents’ Professor and a Georgia Power Distinguished Professor in the School standardized methods for evaluating PV PCS’s and for developing a laboratory
of Electrical Engineering. He is the Founding Director of the University Center that supports PV PCS manufacturers. Dr. Bonn has over 30 years’ experience in
of Excellence for Photovoltaic Research and Education at Georgia Tech. His re- instrumentation and measurements.
search interests include design and fabrication of high-efficiency silicon solar
cells; performance and reliability of PV systems; silicon MOS devices; semi-
conductor material and device characterization; defects and recombination in
semiconductors; and MBE and MOCVD growth of compound semiconductors.

Gregory A. Kern is the Manager of Research and Development at Ascension

Technology’s Boulder, CO lab. Mr. Kern has been responsible for the develop-
ment of Ascension Technology’s AC PV modules, the SunSine™300 and the S. Gonzalez obtained the M.S.E.E. from New Mexico State University. He has
SunSine™325 under development. He has conducted extensive research into been at Sandia National Laboratories for 2 years as an Electrical Engineer on
anti-islanding issues for UIPV systems. the Technical Staff. His responsibilities include instrumentation and test design.