IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS

Online Two-Section PV Array Fault Diagnosis

with Optimized Voltage Sensor Locations
Yihua Hu, Member, IEEE, Jiangfeng Zhang, Wenping Cao, Senior Member, IEEE, Jiande Wu, Member, IEEE,
Gui Yun Tian, Senior Member, IEEE, Stephen J. Finney, James L. Kirtley, Life Fellow, IEEE

 example, a 20 kW grid-connected PV system generally employs

Abstract—Photovoltaic (PV) stations have been built widely 80 modules to form a 20×4 array (i.e. 20 modules in a string
in the world to utilize solar energy directly. In order to reduce the and 4 strings to form an array).
capital and operational costs, early fault diagnosis is playing an In field conditions, a number of factors can cause the PV
increasingly important role by enabling the long effective
operation of PV arrays. This paper analyzes the terminal
array to reduce its output power. In this paper, any cause for this
characteristics of faulty PV strings and arrays, and develops a PV reduction is considered as the “fault”. It can be permanent (such
array fault diagnosis technique. The terminal current-voltage as open-circuits, short-circuits and device aging), or temporary
curve of a faulty PV array is divided into two sections: a high- (such as dust, leave, bird dropping and shadow). A temporary
voltage and a low-voltage fault diagnosis section. The fault can be cleared after a short period of time while a
corresponding working points of healthy string modules, healthy permanent fault would persist over time. Temporary faults can
and faulty modules in an unhealthy string are then analyzed for normally be identified by human eyes and thus be cleared
each section. By probing into different working points, a faulty PV through maintenance. Some permanent faults can be seen if the
module can be located. The fault information is of critical damage is severe while other permanent faults are invisible to
importance for the maximum power point tracking (MPPT) and
the array dynamical reconfiguration. Furthermore, the string
the naked eye so that they may propagate and cause the PV
current sensors can be eliminated while the number of voltage modules to deteriorate over time. PV faults can occur in the PV
sensors can also be reduced by optimizing voltage sensor locations. array and generate different effects on the performance and
Typical fault scenarios including mono-string, multi-string and lifetime of the PV system [11]-[16]. Currently, thermal cameras
partial shadow for a 1.6 kW 3×3 PV array are presented and [17]-[23], earth capacitance measurements (ECM) [24] and
experimentally tested to confirm the effectiveness of the proposed time domain reflectomery (TDR) [25] are the three popular
fault diagnosis method. methods for PV fault diagnosis. Thermal cameras can be
employed to detect the temperature characteristics of a PV array
Index Terms—Fault diagnosis, optimization, photovoltaics, under fault conditions. Thermal images can also be linked to the
terminal characteristics, voltage sensors.
maximum power point tracking (MPPT) algorithm of the PV
controller [22]. In practice, a gradual change in the thermal
I. INTRODUCTION image of the PV module (e.g. due to device aging) poses a
HOTOVOLTAIC (PV) systems provide a promising solution technical challenge [23], and high system costs also limit the
P to directly utilizing solar energy and are currently gaining
in popularity as the technologies are mature and the material
wide application of thermal cameras. The ECM can locate the
disconnection of PV strings while the TDR technology can
costs are driven down [1]-[5]. However, as they are installed in predict the degradation of the PV array. Nonetheless, both ECM
outdoor environments, operational and maintenance costs have and TDR can only operate offline [24],[25]. In practice, online
always been an issue, demanding some fault diagnosis diagnosis methods are highly desired, which can take
functions to improve system reliability. measurements while the tested device is in operation. To
A PV module consists of dozens of PV cells in series improve this, an automatic supervision and fault detection is
connections. A large number of PV modules connected in series proposed in [26],[27] based on power loss analysis. However,
form a PV string, which can be further connected in parallel to it requires surrounding environmental information and cannot
form a PV array. PV modules are characterized with low power identify the faulty module. An operating voltage-window is
density and low output voltage [6]-[8]. If the PV system is then developed based on the PV string operation voltage and
connected to a power grid, a large number of PV modules are ambient temperature [28]. It can locate the open and short faults
needed to connect in series to achieve a high voltage level. but still cannot identify the faulty module from the array.
Typically, a 400 V bus voltage is required for a 220 V 50 Hz Currently, both offline and online fault diagnosis methods have
single-phase grid, and a 600 V bus voltage for a three-phase been developed. Offline diagnosis methods cannot give real-
grid. Similarly, a large number of PV strings are also needed to time fault information that is the key factor for PV array
connect in parallel to increase their power level [9],[10]. For optimization operation under fault condition. Current, online

Manuscript received October 30, 2014; revised January 24, 2015; April 18, W. Cao and J. L. Kirtley are with the Department of Electrical Engineering
2015 and May 23, 2015; accepted June 7, 2015. and Computer Science, Massachusetts Institute of Technology, Cambridge,
Copyright (c) 2015 IEEE. Personal use of this material is permitted. MA, 02139, U.S.A.
However, permission to use this material for any other purposes must be J. Wu is with the College of Electrical Engineering, Zhejiang University,
obtained from the IEEE by sending a request to pubs-permissions@ieee.org. Hangzhou, 310027, P. R. China.
Y. Hu, J. Zhang and S. Finney are with the Department of Electronic & G. Tian is with the School of Electrical and Electronic Engineering,
Electrical Engineering, University of Strathclyde, Glasgow, G1 1XQ, U.K. Newcastle University, Newcastle upon Type, NE1 7RU, U.K.

1
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
fault diagnosis methods suffer from high costs or incapability experimental results of the faulty cell-unit that includes 24 PV
of locating fault modules. A model-based reconfiguration cells with one faulty PV cell; the experimental environment
algorithm is developed in [29] to realize the fault-tolerant parameters are 790 W/m2 at 24oC. The faulty cell is equivalent
operation. But it needs a large number of electrical relays to to a resistance. As the current increases, the corresponding cell-
reconfigure PV arrays. A similar technology, the in-situ unit output power is decreased dramatically. For instance, the
rearrangement strategy, can decrease the influence of shadow faulty cell-unit works at 0.96 A, and its output power is 4.75 W
[30]-[33]. However, its success depends on three conditions: i) (about 10% of the output at healthy condition) and this power
a large number of relays are used. ii) the health state of all PV reduces to nearly zero when the cell-unit current is higher than
modules should be monitored. iii) high computing resource of 1 A. In order to achieve a global MPP for the PV array, the
the controller is required to calculate complex optimal current is much higher than 1 A under the condition in Fig. 2.
arrangements. These increase the system cost and control Therefore, the output voltage for a faulty cell-unit is effectively
complexity. Paper [34] develops an improved strategy which negligible, as shown in Fig. 2.
combines power channels and relays to combat the shadow Therefore, when a PV module is subjected to partial shading,
influence but it also needs the healthy state of PV modules. its terminal output voltage is lower than the healthy module but
Paper [35] proposes a fingerprint curve of the PV array under higher than zero. In Fig. 2(b), the PV module loses one of the
shading conditions to find the key information (e.g. open-circuit 𝑛−1
cell-units and its output voltage is reduced to of the output
𝑛
and short-circuit points and MPP region) but it cannot locate the
voltage.
faulted modules. Paper [36] presents a fault diagnosis technique
PV string fault diagnosis can be achieved by measuring the
using current and voltage sensors but the system cost is quite
PV module voltage, which changes with the string working
high. Paper [22] presents a method to use the fault diagnosis
point. When the string works in the low voltage diagnosis
information for global MPPT without a need to trace I-V curves.
section, the faulty module can be located because its output
It becomes clear that online fault diagnosis is important because
voltage is zero (full shadow) or lower than the healthy module
i) it is the prerequisite for any array dynamical reconfiguration.
(partial shadow).
ii) it can provide crucial information for global MPPT; (iii) it
contains key state-of-health information useful for system B. PV array faults
maintenance. When a PV array is faulted, the faulty module has a lower
This paper proposes a low cost and online fault diagnosis effective illumination than healthy modules. Take a 3×3 array
method with optimized voltage sensor locations that can for example. Fig. 3(a) shows a multi-string faulty condition and
effectively locate the faulty PV strings and faulty modules. The Fig. 3(b) shows its I-V characteristics. In Fig. 3(a), the diodes
paper is organized as follows. Section II introduces PV fault are used to block the reverse current when a fault occurs. The
mechanisms. Section III illustrates the optimization of sensor output I-V characteristics can be divided into two sections: a
locations. Section IV describes the two-section PV array fault high voltage diagnosis section and a low voltage diagnosis
diagnosis method. Section V presents experimental results to section (constant output current). In the latter section, the faulty
verify the proposed method, followed by a short conclusion in module in the faulty string is shorted by bypass diodes where
Section VI. both healthy string and unhealthy string carry the same current.
PV string current sensors cannot distinguish the unhealthy
II. FAULT MECHANISMS string from healthy strings. Nevertheless, the healthy modules
in the faulty string have a higher output voltage than the
Firstly, it is crucial to understand fault mechanisms prior to modules in the healthy string, as points A1 and A2 illustrated in
developing fault diagnosis techniques. Fig. 3(c). The voltage difference between the healthy module in
A. PV string faults the unhealthy string, and the module in the healthy string can
The PV string is the basic structure of a PV array. Fig. 1 be employed to locate the faulty module.
presents typical output characteristics of the PV string under
faulty conditions; the PV module parameters are listed in Table
I. The string includes three modules with non-uniform
illumination, the corresponding environment parameters are
850 W/m2, 25oC; 620 W/m2, 25oC; 400 W/m2, 25oC. Each
module has uniform illumination. It can be found that: i) The
multi-stage characteristics are caused by the differing output
current of each module; ii) In the low voltage diagnosis section,
the faulty modules are short-circuited, and the terminal voltage
of the corresponding faulty module is zero.
Fig. 1 Output characteristics of the faulty string.
In order to restrict the hot-spots in a PV module, a bypass TABLE I SPECIFICATIONS OF THE PV MODULE
diode is connected in parallel to PV cells. The corresponding Parameter Value
structure is named the cell-unit, which is composed of m PV Open-circuit voltage 44.8 V
cells. The PV module is connected in series by n cell units to Short-circuit current 5.29 A
achieve the high output voltage. Usually, partial shadow is also Power output 180 W
MPP current 5A
accrued in one PV module. Due to the cell-unit structure, even MPP voltage 36 V
though only one cell is faulty (0 W/m2), the output power of the Current temperature coefficient 0.037%/K
cell-unit will decrease dramatically. Fig. 2(a) presents Voltage temperature coefficient 4

2
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
Power temperature coefficient 8 information.
Operating cell temperature 46±2°C

(a)
(a) 18

Array Maximum power point

14 Healthy string

Istring/Iarray (A)
Faulty string 1
Faulty string 2
10
Low voltage fault
High voltage fault
section
section
(constant current)
6

0 40 80 120
Uarray (V)
(b) (b)
Fig. 2 PV string under partial shading conditions. (a) The PV cell-unit
output. (b) Partial shading illustration.

In this paper, s modules are connected in series to form a PV

string and p PV strings are connected in parallel to form a PV
array. For a p row s column array, assume that there are x
faulted modules in the unhealthy string. In Fig. 3(c), UA1 is the
voltage of a PV module in a healthy string, such as PV11; UA2
is the voltage of a healthy PV module in an unhealthy string,
such as PV22. UA1 and UA2 can be expressed as: (c) (d)
Fig. 3 PV array under fault conditions. (a) Faulty 3×3 PV array. (b)
U
U A1  array Output characteristics. (c) Working points in the voltage fault
s (1) diagnosis section. (d) Working points in the high-voltage section.
U array
U A2  The extreme condition for the PV array under non-uniform
sx (2)
illumination is that the illumination on the faulty module is zero.
where Uarray is the output voltage of PV arrays. Fig. 4 presents the output curves of healthy string and unhealthy
The high voltage diagnosis section in Fig. 3(b) is due to a string under this condition. There is a zero-output condition,
lower solar illumination of the faulty module. The output where the faulty string does not generate electricity. When the
current of the unhealthy string is limited by the faulty module array output voltage is between 100~130 V, every module in
output current. Therefore, the unhealthy string output current is the healthy string generates electricity and works in the high-
lower than the healthy string. Since all the modules contribute
voltage diagnosis section. In the unhealthy string, the faulty
to electricity generation, there are three working points in two
module cannot generate electricity. Although the healthy
output characteristics. A3 is the working point of modules in the
modules work in the high-voltage diagnosis section, the
healthy string; A4 is working point of the faulty modules in the
unhealthy string; A5 is the working point of normal modules in unhealthy string still cannot reach the PV array voltage.
the unhealthy string; as in Fig. 3(d). Because both A3 and A5 are Therefore the healthy modules in the unhealthy string are
the working points of a healthy module, they share the same effectively open-circuited, similar to the faulted modules. There
output curve characteristics. Because A4 and A5 are the working is neither current flowing in the unhealthy string, nor in the
points of an unhealthy module and a healthy module in the same bypass diodes. That is, all the modules in an unhealthy string
string, they have the same output current. are open-circuited.
Voltages UA3, UA4 and UA5 for working points A3, A4 and A5 Therefore, the fault diagnosis can be achieved by analyzing
are given by: the module voltage at different diagnosis sections. This also
U A3  s  U A4  x  U A5  (s  x) removes the necessity of current sensors. In this work, the two-
(3) stage power conversion [37] is adopted so that the control of the
According to the previous analysis, UA4＜UA3＜UA5. This can PV system is load independent. That is, the PV’s working point
be employed to locate faulty modules without current can be chosen at will in the two-stage PV system where the

3
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
front-end DC-DC converter tracks the Fig. 5 Voltage sensor placement methods: (a) 1, (b) 2, (c) 3.

Fig. 6 Equivalent matrix.

Fig. 4 Extreme case of the PV array fault.

desired working point of the PV array; the bus voltage control

V Ua
and the DC-AC inverter control ensure that the grid current is
controlled as per the input power.

III. OPTIMIZATION OF SENSOR LOCATIONS

Ub V V Uc
In order to achieve the PV array fault diagnosis, the reading
of PV module voltage is needed. Due to the large number of PV
Fig. 7 Simplified voltage sensor placement for a 33 PV array.
modules employed, a large number of voltage sensors are also
needed in the first instance. TABLE II NUMBER OF VOLTAGE SENSORS USED BY DIFFERENT METHODS
A. Sensor placement strategy Method 1 2 3 Proposed
No. p s (p-1)(s-1) p(s-2) p(s-1)/2
There are three basic sensor placement methods, as shown in
Fig. 5. If every module’s terminal voltage is measured by a
Fig. 7 presents an example of the 3×3 PV array. According
voltage sensor by method 1; and the total number of sensors is
to the proposed sensor placement strategy, only three voltage
ps. In method 2, each voltage sensor measures the voltage
sensors are needed.
between two nodes in the same column of adjacent strings; and
The minimum number of sensors used to detect all possible
(p1)(s1) voltage sensors are needed. In method 3, the faults should be ⌈𝑝 × (𝑠 − 1)/2⌉. When a node is not connected
electric potential difference of adjacent modules is measured; to any sensor, the two adjacent PV modules of this node cannot
the corresponding number of sensors is p(s2). The large be discriminated once a fault occurs at one of the two modules.
number of voltage sensors may increase system capital cost and The total number of nodes is equal to ⌈𝑝 × (𝑠 − 1)/2⌉. This is
information processing burden. Therefore, the voltage summarized in Table II. Clearly, the proposed method uses less
placement method needs to be optimized. voltage sensors than other three methods.
Fig. 6 shows an equivalent PV matrix where a PV module is
shown as a dot; the connection line of the adjacent module is B. Mathematical model of the proposed strategy
represented by a node. The proposed voltage placement strategy The variable 𝑎𝑖𝑗 is defined as the state of the PV module
is developed by the following steps: sitting at the i-th string and j-th module (i.e. (𝑖, 𝑗)) in the 𝑝 × 𝑠
i) All the nodes should be covered by voltage sensors. array. If this module is healthy, 𝑎𝑖𝑗 =1, otherwise 𝑎𝑖𝑗 =0. The
ii) A sensor can only connect one node in a string. terminal voltage of the (i, j) module is denoted by 𝑢𝑖𝑗 , and the
iii) Voltage sensor nodes cover different isoelectric points reading of a voltage sensor connecting the (i, j) and (r, k)
from different strings. modules is denoted by 𝑅𝑖,𝑗,𝑟,𝑘 . Consider that each string has at
iv) If p or s is an even number, each node is connected to
least one healthy module. The number of healthy modules in
and only to one sensor. If both p and s are odd, there is the i-th string equalizes 𝑎𝑖1 + 𝑎𝑖2 + ⋯ + 𝑎𝑖𝑠 . The terminal
one and only one node to be connected to two different voltage 𝑢𝑖𝑗 of the (i, j) module is equal to a fraction of 𝑈𝑎𝑟𝑟𝑎𝑦 ,
sensors, while each of remaining nodes is connected to
and this fraction is 0 if 𝑎𝑖𝑗 = 0, and is 1/( 𝑎𝑖1 + 𝑎𝑖2 + ⋯ +
one sensor.
𝑎𝑖𝑠 ) if 𝑎𝑖𝑗 = 1. That is,
𝑎𝑖𝑗 𝑈array
V V V
V V
𝑢𝑖𝑗 = (4)
𝑎 +𝑎 +⋯+𝑎
𝑖1 𝑖2 𝑖𝑠
Note that the total output voltage of the modules (i, 1), (i,
V V V V V V V V
2),…, and (i, j) is the sum of the terminal voltage of j modules,
i.e., 𝑢𝑖1 + 𝑢𝑖2 + ⋯ + 𝑢𝑖𝑗 . Similarly the total output voltage of
the modules (r, 1), (r, 2),…, and (r, k) equalizes 𝑢𝑟1 + 𝑢𝑟2 +
V V V.
5 V V V
V V
⋯ 𝑢𝑟𝑘 . Therefore, the reading of the voltage sensor connecting
the (i, j) module and the (r, k) module is calculated as
𝑅𝑖,𝑗,𝑟,𝑘 = (𝑢𝑖1 + 𝑢𝑖2 + ⋯ + 𝑢𝑖𝑗 ) − (𝑢𝑟1 + 𝑢𝑟2 + ⋯ 𝑢𝑟𝑘 )
(𝑎𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑗 )𝑈array (𝑎𝑟1 +𝑎𝑟2 +⋯+𝑎𝑟𝑘 )𝑈array
(a) (b) (c) = − (5)
𝑎𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑠 𝑎𝑟1 +𝑎𝑟2 +⋯+𝑎𝑟𝑠

4
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
When the working point of a PV string moves to the high the first k modules in the r-th string (i.e. 𝑢𝑟1 + 𝑢𝑟2 +
voltage section, the output voltage of the healthy modules ⋯ 𝑢𝑟𝑘 ). Whenever there is current flowing in the r-th
increases until reaching 𝑈𝑜𝑐 . The faulted modules in the string string, there will be at least one module works at low
will equally divide the remaining voltage 𝑈array −( 𝑎𝑖1 + 𝑎𝑖2 + voltage working points (e.g., 𝑈array < 𝑈𝑜𝑐 ). At the low
⋯ + 𝑎𝑖𝑠 )𝑈𝑜𝑐 . The following relations hold for a string including voltage section, the reading of 𝑢𝑟1 + 𝑢𝑟2 + ⋯ 𝑢𝑟𝑘 is a
both healthy and unhealthy modules. function of 𝑈array and cannot remain constant.
(1−𝑎𝑖𝑗 )(𝑈array −(𝑎𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑠 )𝑈𝑜𝑐 )
𝑢𝑖𝑗 = 𝑎𝑖𝑗 𝑈𝑜𝑐 + 𝑠−(𝑎𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑠 )
B. Locating faulty PV modules in the low-voltage section
(𝑎𝑖𝑗 𝑠−(𝑎𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑠 ))𝑈𝑜𝑐 (1−𝑎𝑖𝑗 )𝑈array After locating the healthy string, the next step is to find the
= 𝑠−(𝑎𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑠 )
+ 𝑠−(𝑎 (6) faulty PV module. In the low voltage diagnosis section, the
𝑖1 +𝑎𝑖2 +⋯+𝑎𝑖𝑠 )
𝑅𝑖,𝑗,𝑟,𝑘 = (𝑢𝑖1 + 𝑢𝑖2 + ⋯ + 𝑢𝑖𝑗 ) − (𝑢𝑟1 + 𝑢𝑟2 + ⋯ 𝑢𝑟𝑘 ) faulty modules are shorted. The corresponding fault diagnosis
𝑗 𝑗 eigenvalue of the mono-string faulty is presented in Table III,
(𝑠 ∑𝑙=1 𝑎𝑖𝑙 − 𝑗 ∑𝑠𝑙=1 𝑎𝑖𝑙 )𝑈𝑜𝑐 (𝑗 − ∑𝑙=1 𝑎𝑖𝑙 )𝑈array
= + where the fully-faulty module indicates that all cell-units in the
𝑠 − ∑𝑠𝑙=1 𝑎𝑖𝑙 ∑𝑠𝑙=1 𝑎𝑖𝑙 module are faulty (0: healthy and 1: faulty). No. 7 (111) is the
(𝑠 ∑𝑘 𝑠
𝑙=1 𝑎𝑟𝑙 −𝑘 ∑𝑙=1 𝑎𝑟𝑙 )𝑈𝑜𝑐 (𝑘−∑𝑘
𝑙=1 𝑎𝑟𝑙 )𝑈array extreme case that all the modules in this string are faulty. Even
− 𝑠−∑𝑠𝑙=1 𝑎𝑟𝑙
− ∑𝑠𝑙=1 𝑎𝑟𝑙
(7) though the PV array works in the low-voltage diagnosis section,
The reading 𝑅𝑖,𝑗,𝑟,𝑘 at the high voltage section provides the modules are open-circuited when all modules are faulty.
extra equations to solve variable 𝑎𝑖𝑗 . There is a way to design Table IV shows the multi-string eigenvalues. From these, the
the optimal sensor placement for any 𝑝 × 𝑠 array with faulty module can be identified easily.
⌈𝑝 × (𝑠 − 1)/2⌉ sensors. If p is an even number, the 𝑝 × 𝑠
𝑝 TABLE III VOLTAGE OF THE MONO-STRING ALL FAULTED MODULES
array can be divided into elements of 2 × 𝑠 arrays. For each PV31~PV33 Ua Ub Uc
2
2 × 𝑠 array, it needs to apply the optimal sensor placement 100 Uarray/3 2Uarray/3 Uarray/6
𝑝 010 Uarray/3 Uarray/6 Uarray/6
method by using × 𝑠 sensors. If p is odd, the 𝑝 × 𝑠 array 001 Uarray/3 Uarray/6 2Uarray/3
2
𝑝−3 110 Uarray/3 2Uarray/3 Uarray/3
consists of one 3 × 𝑠 array and elements of 2 × 𝑠 arrays. It
2 011 Uarray/3 Uarray/3 2Uarray/3
needs to apply the sensor placement method for these elements 101 Uarray/3 2Uarray/3 2Uarray/3
and the number of sensors needed is equal to ⌈3 × (𝑠 − 1)/2⌉ + 111 Uarray/3 2Uarray/3-Uoc 2Uoc -Uarray/3
𝑝−3 000 Uarray/3 Uarray/3 Uarray/3
(𝑠 − 1). By considering both even and odd numbers,
2
𝑝−3 TABLE IV VOLTAGE OF THE MULTI-STRINGS FULLY FAULTED MODULES
⌈3 × (𝑠 − 1)/2⌉ + (𝑠 − 1) = ⌈𝑝 × (𝑠 − 1)/2⌉ (8) PV11~PV13/PV21~PV23 Ua Ub Uc
2
Therefore, the optimal number of sensors can be obtained. 100/100 Uarray/2 Uarray/6 2Uarray/3
010/100 Uarray/2 Uarray/6 Uarray/6
001/100 Uarray Uarray/6 Uarray/6
IV. TWO-SECTION PV ARRAY FAULT DIAGNOSIS STRATEGY 100/010 Uarray/6 Uarray/6 2Uarray/3
The proposed PV array fault diagnosis strategy is 010/010 Uarray/6 Uarray/6 Uarray/6
001/010 Uarray/2 Uarray/6 Uarray/6
implemented in three steps: locating healthy PV string, locating 100/001 0 2Uarray/3 2Uarray/3
faulty module in the low-voltage diagnosis section, and in the 010/001 0 2Uarray/3 Uarray/6
high-voltage diagnosis section. 001/001 Uarray/2 2Uarray/3 Uarray/6
110/100 0 Uarray/6 2Uarray/3
A. Locating healthy PV strings 101/100 Uarray Uarray/6 2Uarray/3
011/100 Uarray Uarray/6 Uarray/3
The information of healthy strings is useful to identify a
faulty module. Thus the first step in fault diagnosis is to locate In practice, partial shading is a very common fault [2]-
healthy PV strings. Because of the absence of current sensors [3][5][12]-[14][22]-[33]. This is illustrated in detail in Table V.
in the string, the healthy string cannot be found directly. When Both Tables III and V are concerned with PV module faults.
a PV array changes from a healthy condition to an unhealthy Tables III deals with the fully-faulted module where all cell-
condition, the voltage sensor can pick up the change. units are faulted while Table V shows a partially faulted module
i) If the voltage sensor reading 𝑅𝑖,𝑗,𝑟,𝑘 always satisfies including some faulted cell-units. Their output voltages are zero
𝑗−𝑘
𝑅𝑖,𝑗,𝑟,𝑘 = 𝑈array despite any changes of the working and non-zero, respectively.
𝑠
point along the I-V curve, both i-th and r-th strings are C. Locating faulty PV module in the high-voltage section
healthy.
If all the PV strings are faulty, the eigenvalues of Tables III-
ii) If the i-th string is healthy, the sensor reading 𝑅𝑖,𝑗,𝑟,𝑘
𝑅𝑖,𝑗,𝑟,𝑘 𝑗 𝑎𝑟1 +𝑎𝑟2 +⋯+𝑎𝑟𝑘
V may be the same as other faulty conditions. This can lead to
satisfies − =− at low voltage misjudgment in locating faulty modules.
𝑈array 𝑠 𝑎𝑟1 +𝑎𝑟2 +⋯+𝑎𝑟𝑠
working points. This can be used to judge the number of For example, two types of the unhealthy 3×3 PV array with
faulty modules in the r-th string. the same sensor placement strategy are presented in Fig. 8.
𝑗 PV11, PV21, PV23 and PV32 are faulty at fault condition 1;
iii) If the i-th string is healthy, and (𝑅𝑖,𝑗,𝑟,𝑘 − 𝑈array ) remains
𝑠 PV12, PV21, PV31 and PV33 are faulty at condition 2. Two
constant for all working points, there is no current flowing fault conditions give the same voltage reading in the low-
in the r-th string, i.e., the r-th string is open circuited. This voltage diagnosis section, which is Uarray/2. In order to
𝑗
is because that (𝑅𝑖,𝑗,𝑟,𝑘 − 𝑈array ) is equal to the voltage of discriminate the two conditions, the high voltage diagnosis
𝑠
section is employed to find the actual faulty modules.
5
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
TABLE V VOLT OF THE MONO-STRING PARTIALLY FAULTED MODULES Monitor values from
PV11~PV13 Ua Ub Uc Comparison voltage sensor

Uarray/6<Ua< Uarray/3<Uc<
100 Uarray/3 —
Uarray/3 2Uarray/3
Uarray/6<Ua< Uarray/6<Uc<
010 Uarray/3 — Are all values normal?
Y
Uarray/3 Uarray/3
Uarray/3<Ua< Uarray/6<Uc<
001 Uarray/3 —
2Uarray/3 Uarray/3
Uarray/3<Uc<
110 Ua<Uarray/3 Uarray/3 2Ua Uc< Uarray
2Uarray/3 Is there a healthy
string? N
Uarray/3<Ua<
011 Uarray/3 Uc<Uarray/3 Ua 2Uc< Uarray
2Uarray/3 Y
Uarray/3<Uc< Low voltage area Low voltage area High voltage area
101 Ua<Uarray/3 Uarray/3 Ua 2Uc> Uarray
2Uarray/3 fault diagnosis fault diagnosis fault diagnosis

Locate faulty Locate faulty

module module

Fig. 9 Flowchart of the two-section fault diagnosis.

processed by the conditioning circuit and then input to DSP

TMS320F2812. The PV modules are the same for simulation,
and the environment illumination is recorded by TS1333R. In
the experiment, typical fault scenarios are studied and the
(a) (b)
sensor readings are compared with eigenvalues in the high-
Fig. 8 The 3×3 array under two fault conditions: (a) 1, (b) 2.
voltage and low-voltage diagnosis sections to check the
TABLE VI EIGENVALUE UNDER DIFFERENT FAULT DIAGNOSIS SECTIONS effectiveness of the proposed fault diagnosis technique.
Fault condition Diagnosis section Ua Ub Uc Fig. 11 shows the mono-string mono-module fault diagnosis.
1 Low voltage Uarray/2 Uarray/2 Uarray/2 In the fault scenario 1 (see Fig. 11(a)), the voltage sensor
2 Low voltage Uarray/2 Uarray/2 Uarray/2
1 High voltage Uarray/2-Uoc 2Uoc-Uarray/2 Uarray/2
connection method is identical to that in Fig. 7. The
2 High voltage Uarray/2 Uarray/2-Uoc 2Uoc-Uarray/2 illumination is 550 W/m2 and the temperature is 15℃. The P33
PV module is cast by shadow manually to emulate a partial-
As analyzed previously, when the PV array works in the high shading fault. Uaref, Ubref and Ucref are the reference voltages for
diagnosis voltage section, the faulty modules in string 2 are sensors a, b and c, respectively, under the fault condition. Fig.
open-circuited while strings 1 and 3 can still operate. In a low- 11(b) shows the I-V characteristics of faulty PV arrays. Due to
voltage section all the healthy modules in the array generate the fault on module P33, string 3 cannot generate electricity in
electricity and the faulty modules are shorted. The voltage value the output voltage range 82~120 V. Fig. 11(c) presents the
is different between the low-voltage and the high voltage sensor output voltage. In the low-voltage diagnosis section
sections, as illustrated in Table VI. By changing the diagnosis (10~70V); the sensor a output voltage is Uarray/3 (as shown in
section from low to high, different eigenvalues can be obtained Fig. 11(d)). This is a normal output voltage and the
to locate faulty modules. corresponding strings are healthy. That is, strings 1 and 2
connected by this sensor are healthy, which coincides with fault
D. Implementation of the two-section fault diagnosis strategy scenario 1 in Fig. 11(a). Fig. 11(e) illustrates the high-voltage
The two-section fault diagnosis strategy is summarized in a and low-voltage diagnosis sections. In the low-voltage section,
flowchart in Fig. 9. Firstly, the system checks if the voltage the reference eigenvalue is Uarray/6; and in the high-voltage
sensor readings sufficiently deviate from the normal ones (e.g. section, it is 2Uarray/3-Uoc. The fact that the sensor b output is
10%). If this is true, the system enters into fault diagnosis. Next, close to the reference value also verifies the proposed diagnosis
the system checks if there is a healthy string in the PV array method. The reference eigenvalue of Uc is 2Uarray/3; and the
based on voltage sensor readings. If this is the case, the faulty corresponding sensor c output also agrees with the reference
module can be located directly by the eigenvalue table in the eigenvalues. There is a slight deviation between Ua, Ub and Uc
low-voltage diagnosis section. If there does not exist a healthy and their reference values. This is caused by the diode voltage
string, both the low-voltage and high-voltage fault diagnosis is drop and the minor product irregularity between PV modules.
needed to locate faulty modules. From the sensor output results and information in Table III, the
fault type is classified as “001”. The faulty module is P33 that
V. EXPERIMENTAL TESTS also agrees with fault scenario 1 (Fig. 11(a)).
A 3×3 PV array and a signal conditioning system are built Fig. 12 shows the diagnosis of the multi-string mono-module
to verify the proposed fault diagnosis technique, as shown in fault. In fault scenario 2, the illumination is 580 W/m2 and the
Fig. 10. In this figure, resistor dividers are employed as temperature is 25℃. Module P11 in string 1 and P33 in string
differential voltage sensors to eliminate the grounding issue, tri- 3 are cast by partial shadow manually. Fig. 12(b) presents the I-
port connectors and shield twisted pair cables are used to V characteristics of the faulty PV array. When faults occur in
transmit the voltage signal. In the signal conditioning circuit, modules P11 and P33, strings 1 and 3 cannot generate
the electrical isolation of sensor signals is achieved by using a electricity in the range of 82~120 V. In the low-voltage section,
linear optical coupling (HCNR201) to avoid the interaction of the sensors a and b have the same output (Ua=Ub=Uarray/6), as
earth and ground connections. The voltage readings are illustrated in Fig.12(c). The voltage sensors a and b also satisfy

6
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
the rule for locating healthy strings. Therefore, string 2 is
diagnosed as being healthy which coincides with the fault
scenario in Fig. 12(a). Fig. 12(d) shows the sensor c output
curves. It can be seen that there is a healthy string, and the
values of Ua, Ub and Uc are Uarray/6, Uarray/6 and Uarray,
respectively, in the low-voltage diagnosis section. The faulty
modules identified are P11 and P33.

Tri-port
connector
(a) (b)

PV array

Shield twisted pair cable

(c) (d)
Fig. 12 Multi-string mono-module fault diagnosis. (a) Fault scenario 2.
(b) Output characteristics. (c) Comparison of sensor outputs Ua and Ub.
signal (d) Sensor output voltage.
DSP
conditioning
2812
circuit
Resistor
divider

Two stage
converter

(a)
Fig. 10 Experimental platform.

(b) (c)
(a) (b)

(c) (d) (d) (e)

Fig. 13 Mono-string multi-module fault diagnosis. (a) Fault scenario 3.
(b) Output characteristics. (c) Ua sensor output. (d) Comparison of
sensor output Uc. (e) Comparison of sensor output Ub.

In fault scenario 3 (Fig. 13(a)), the illumination is 610 W/m2

at 30℃. P32 and P33 in string 3 are cast by partial shadow and
full shadow, respectively. Fig. 13(b) shows the output
characteristics. String 3 can only generate electricity at 0~60 V.
(e) (f) As presented in Fig. 13(c), Ua =Uarray/3 in the whole output
Fig. 11 Mono-string mono-module fully faulted diagnosis. (a) Fault voltage range, indicating strings 1 and 2 are both healthy. In the
scenario 1. (b) Output characteristics. (c) Voltage sensor output. (d) Ua low-voltage section of Fig. 13(d), Uc matches the reference
sensor output. (e) Comparison of Ub sensor output. (f) Comparison of
value 2Uarray/3, verifying that P33 is faulty. In the high-voltage
Uc sensor output.
section, Uc= 2Uoc-Uarray/3. Therefore, either P31 or P32 is faulty.
Fig. 13(e) presents the output characteristics for sensor b. The
7
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
corresponding output is equal to -Uarray/3 in the low-voltage REFERENCES
range, proving that P32 module is faulty and P31 module is [1] Y. A. Mahmoud, W. Xiao, H. H. Zeineldin, “A parameterization approach
healthy. for enhancing PV model accuracy,” IEEE Trans. Ind. Electron., vol. 60,
From the analysis of three fault scenarios for a 3×3 PV array, no.12, pp. 5708-5716, 2013.
[2] B. N. Alajmi, K. H. Ahmed, S. J. Finney, B. W. Williams, “A maximum
the proposed fault diagnosis strategy is proven be effective. power point tracking technique for partially shaded photovoltaic systems
For small-scale or low-voltage arrays, passive voltage in microgrids,” IEEE Trans. Ind. Electron., vol. 60, no. 4, pp. 1596-1606,
sensors (e.g. resistor dividers) can be employed. Given that 16 April 2013.
channels are available for analog-digital (A/D) conversion in [3] L. Gao, R. A. Dougal, S. Liu, and A. P. Iotova, “Parallel-connected solar
PV System to address partial and rapidly fluctuating shadow conditions,”
DSP TMS320F28335, there might not need for additional A/D IEEE Trans. Ind. Electron., vol. 56, no. 5, pp. 1548-1556, May 2009.
chips. For large-scale PV systems at high voltages, Hall-effect [4] N Femia, G Petrone, G Spagnuolo, M Vitelli,” A technique for improving
voltage sensors (e.g. LEM LV25-P) are required. These sensors P&O MPPT performances of double-stage grid-connected photovoltaic
can be powered by the PV cell-unit directly. They also need systems,” IEEE Trans. Ind. Electron., vol. 56, no. 11, pp. 4473-4482, Nov.
2009.
long sensor cables to transmit the voltage results unless wireless [5] K. Ishaque, and Z. Salam, “A deterministic particle swarm optimization
sensor networks are used [38][39]. Clearly, the cost of the fault maximum power point tracker for photovoltaic system under partial
diagnosis equipment and computational complexity handling shading condition,” IEEE Trans. Ind. Electron., vol. 60, no. 8, pp. 3195-
for more extended PV arrays and large PV arrays will be 3206, Aug. 2013.
[6] M. Boztepe, F. Guinjoan, G. Velasco-Quesada, S. Silvestre, A. Chouder,
increased. Given the gain in reduced voltage sensors and E. Karatepe, “Global MPPT scheme for photovoltaic string inverters based
increased solar power production, the total capital cost is on restricted voltage window search algorithm,” IEEE Trans. Ind.
justified by using the proposed technique. Electron., vol. 61, no. 7, pp. 3302-3312, Jul. 2014.
It needs to point out that this is a proof-of-concept work and [7] K. Ding, X. Bian, H. Liu, T. Peng, “A MATLAB-Simulink-based PV
module model and its application under conditions of nonuniform
its technology readiness level (TRL) is between 3-4. Ideally, the irradiance,” IEEE Trans. Energy Convers., vol. 27, no. 4, pp. 864-872,
developed technology will eventually lead to a new product, in Dec. 2012.
place of existing converters for PV systems. However, it can [8] G Petrone, G Spagnuolo, M Vitelli, “Analytical model of mismatched
also be integrated into the exiting commercial converters. i) If photovoltaic fields by means of Lambert W-function,” Solar Energy
Materials and Solar Cells, vol. 91 no. 18, pp. 1652-1657, Nov. 2007.
commercial converters allow for updating their software [9] Y. Hu, Y. Deng, Q. Liu, X. He, “Asymmetry three-level grid-connected
programs, the developed algorithm can be implemented into the current hysteresis control with varying bus voltage and virtual over-sample
control program of the front-end DC-DC converter and voltage method,” IEEE Trans. Power Electron., vol. 29, no. 6, pp. 3214-3222, Jun.
sensors need to add to the system for voltage measurement. For 2014.
[10] Y. Liu, B. Ge, H. Abu-Rub, and F. Z. Peng, “An effective control method
fault diagnosis, the reference voltages of PV arrays in the low- for three-phase quasi-Z-source cascaded multilevel inverter based grid-tie
voltage and high-voltage fault diagnosis areas are firstly photovoltaic power system,” IEEE Trans. Ind. Electron., vol. 61, no. 12,
chosen. The difference between the PV array output voltage and pp. 6794-6802, Dec. 2014.
the reference voltage is the input to the PI controller and its [11] S. Djordjevic, D. Parlevliet, P. Jennings, “Detectable faults on recently
installed solar modules in Western Australia” Renewable Energy, vol. 67,
output is the duty cycle of the DC-DC converter. ii) If pp. 215-221, 2014.
commercial converters do not allow any modification of their [12] A. Maki, S. Valkealahti, “Effect of photovoltaic generator components on
software programs, an extra DC-DC converter is needed and its the number of MPPs under partial shading conditions,” IEEE Trans.
output is connected across the DC-link capacitor. This Energy Convers., vol. 28, Issue 4, pp.1008-1017, 2013.
[13] M. Z. S El-Dein, M. Kazerani, M. M. A. Salama, “Optimal photovoltaic
arrangement bypasses the first stage of the commercial array reconfiguration to reduce partial shading losses,” IEEE Trans.
converter and fault diagnosis can be conducted when the PV Sustain. Energy, vol. 4, Issue 1, pp. 145-153, 2013.
system is operational. [14] E. V. Paraskevadaki, S. A. Papathanassiou, “Evaluation of MPP voltage
and power of mc-Si PV modules in partial shading conditions,” IEEE
Trans. Energy Convers., vol. 26, Issue 3, pp. 923-932, 2011.
VI. CONCLUSION [15] H. A. Lauffenburger, R. T. Anderson, “Reliability terminology and
PVs are a cost-sensitive market. Online fault diagnosis is formulae for photovoltaic power systems,” IEEE Trans. Rel., vol. R-31,
Issue 3, pp. 289-295, 1982.
key to the success of the PV array reconfiguration and the [16] L. H. Stember, W. R. Huss, M. S. Bridgman, “A methodology for
global MPPT. This paper has proposed a low-cost online PV photovoltaic system reliability & economic analysis,” IEEE Trans. Rel.,
array fault diagnosis with optimized voltage sensor locations. vol. R-31, Issue 3, pp. 296-303, 1982.
This work can increase productivity and reduce the capital and [17] C. Buerhopa, D. Schlegela, M. Niessb, C. Vodermayerb, R. Weißmanna,
C. J. Brabeca, “Reliability of IR-imaging of PV-plants under operating
maintenance costs by reducing the number of sensors and by conditions,” Solar Energy Materials and Solar Cells, vol. 107, pp. 154-
developing an effective fault diagnosis technique. 164, 2012.
Compared to existing methods in the literature, this work [18] Y. Hu, B. Gao, G.Y. Tian, X. Song, K. Li, X. He. “Photovoltaic fault
has made the following improvements: i) String current sensors detection using a parameter based model,” Solar Energy, vol. 96, pp. 96-
102, Oct. 2013.
are removed and the number of voltage sensors is also reduced [19] A. Krenzinger, A. C. Andrade, “Accurate outdoor glass thermographic
by optimizing the location of voltage sensors. ii) An online two- thermometry applied to solar energy devices,” Solar Energy, vol. 81, pp.
section fault diagnosis method is developed to locate faulty PV 1025-1034, 2007.
modules. iii) The state of health information from this work can [20] M. Simon and E. L. Meyer, “Detection and analysis of hot-spot formation
in solar cells,” Solar Energy Materials and Solar Cells, vol. 94, no. 2, pp.
be also used for the MPPT and PV array dynamical 106-113, 2010.
reconfiguration. [21] J. Kurnik, M. Jankovec, K. Brecl and M. Topic, “Outdoor testing of PV
module temperature and performance under different mounting and
operational conditions,” Solar Energy Materials & Solar Cells, vol. 95,
pp. 373-376, 2011.
[22] Y. Hu, W. Cao, J. Wu, B. Ji, D. Holliday, “Thermography-based virtual
MPPT scheme for improving PV energy efficiency at partial shading

8
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
conditions,” IEEE Trans. Power Electron. vol. 29, no. 11, pp. 5667-5672, Jiangfeng Zhang obtained his BSc and PhD in
Jun. 2014. computing mathematics and applied software from
[23] Z. Zou, Y. Hu, B. Gao, W. L.Woo and X. Zhao, “Study of the gradual Xi’an Jiaotong University, China, in July 1995 and
change phenomenon in the infrared image when monitoring photovoltaic December 1999, respectively. He is a senior lecturer at
array,” Journal of Applied Physics, vol. 115, no. 4, pp. 1-11, 2014. the Department of Electronic and Electrical
[24] T. Takashima, J. Yamaguchi, K. Otani, T. Oozeki, K. Kato, “Experimental Engineering, University of Strathclyde, Glasgow, U.K.
studies of fault location in PV module strings,” Solar Energy Materials He is also a member of the IFAC TC6.3 (Power and
and Solar Cells, vol. 93 issues. 6-7, pp. 1079-1082, Jun. 2009. Energy Systems). His research interests include
[25] R. A. Kumar, M. S. Suresh, J. Nagaraju, “Measurement of AC parameters modeling and control of energy systems, with a focus
of gallium arsenide (GaAs/Ge) solar cell by impedance spectroscopy,” on energy efficiency and demand side management.
IEEE Trans. Electron Devices, vol. 48, issue 9, pp. 2177-2179, 2001.
[26] A. Chouder, S. Silvestre, “Automatic supervision and fault detection of
PV systems based on power losses analysis,” Energy Conversion and Wenping Cao (M’05-SM’11) received the B.Eng in
Management, vol. 51, issue 10, pp. 1929-1937, Oct. 2010. electrical engineering from Beijing Jiaotong University,
[27] S. Silvestre, A. Chouder, E. Karatepe, “Automatic fault detection in grid Beijing, China, in 1991, and the Ph.D. degree in
connected PV systems,” Solar Energy, vol. 94, pp. 119-127, Aug. 2013.
electrical machines and drives from the University of
[28] N. Gokmen, E. Karatepe, S. Silvestre, B. Celik, P. Ortega, “An efficient
Nottingham, Nottingham, U.K., in 2004.
fault diagnosis method for PV systems based on operating voltage-
He is currently a Marie Curie Fellow with the
window,” Energy Conversion and Management, vol. 73, pp. 350-360, Sep.
Massachusetts Institute of Technology, Cambridge,
2013.
MA, U.S.A, and a Senior Lecturer with Queen’s
[29] X. Lin, Y. Wang, D. Zhu, N. Chang and M. Pedram, “Online fault detection
University Belfast, Belfast, U.K. Dr. Cao received the
and tolerance for photovoltaic energy harvesting systems,”
“Best Paper Award” at the LDIA’13 Conference, the
the 2012 IEEE/ACM International Conference on Computer-Aided Design
“Innovator of the Year” Award from Newcastle
(ICCAD), San Jose, USA, pp. 1-6. 2012.
University in 2013, and was the Winner of the “Dragons’ Den Competition” at
[30] D. Nguyen, B. Lehman. “An adaptive solar photovoltaic array using
Queen’s University Belfast in 2014. He serves as an Associate Editor for IEEE
model-based reconfiguration algorithm,” IEEE Trans. Ind. Electron., vol.
TRANSACTIONS ON INDUSTRY APPLICATIONS, IEEE INDUSTRY APPLICATIONS
55, no. 7, pp. 2644-2654, Jul. 2008.
MAGAZINE and IET Power Electronics, the Chief Editor for two Special Issues,
[31] J. P. Storey, P. R. Wilson, and D. Bagnall, “Improved optimization strategy
an Editor for Electric Power Components & Systems Journal and nine other
for irradiance equalization in dynamic photovoltaic arrays,” IEEE Trans.
International Journals. His research interests are in fault analysis and condition
Power Electron., vol. 28, no. 6, pp. 2946-2956, Jun. 2013.
monitoring of electric machines and power electronics.
[32] G. Velasco-Quesada, F. Guinjoan-Gispert, R. Pique-Lopez, M. Roman-
Lumbreras, and A. Conesa-Roca, “Electrical PV array reconfiguration
Jiande Wu (M’11) received the B.Sc. and the Ph.D. in
strategy for energy extraction improvement in grid-connected PV
power electronics from Zhejiang University in 1994
systems,” IEEE Trans. Ind. Electron., vol. 56, no. 11, pp. 4319-4331, Nov.
and 2012, respectively. He has been a faculty member
2009.
at Zhejiang University since 1997 and is currently an
[33] Y. Wang, X. Lin, Y. Kim, N. Chang, and M. Pedram, “Architecture and
associate professor. From Oct. 2013 to Oct. 2014, he
control algorithms for combating partial shading in photovoltaic systems,”
was an academic visitor at the University of
IEEE Trans. Comput.-Aided Design Integr. Circuits Syst., vol. 33, no. 6,
Strathclyde, Glasgow, U.K. His research interests
pp. 917-929, Jun. 2014.
include applications of power electronics and network
[34] S. Jonathan, R. W. Peter, and B. Darren, “The optimized-string dynamic
communication.
photovoltaic array,” IEEE Trans. Power Electron., vol. 29, no. 4, pp. 1768-
1776, Apr. 2014.
[35] P. L. Carotenuto, P. Manganiello, G. Petrone, and G. Spagnuolo, “Online
Gui Yun Tian (M’01–SM’03) received the B.Sc. in
recording a PV module fingerprint,” IEEE J. Photovolt, vol. 4, no. 2, pp.
metrology and instrumentation and the M.Sc. in
659-668, Mar. 2014.
precision engineering from the University of Sichuan,
[36] Y. Hu, H. Chen, R. Xu, R. Li. “Photovoltaic (PV) array fault diagnosis
Chengdu, China, in 1985 and 1988, respectively, and
strategy based on optimal sensor placement,” Proceedings of the CSEE,
the Ph.D. degree from the University of Derby, Derby,
vol. 31, issue 33, pp. 19-30, 2011.
U.K., in 1998. He is currently a Chair Professor in
[37] B. Yang, W. Li, Y. Zhao and X. He, “Design and analysis of a grid-
sensor technologies at Newcastle University,
connected photovoltaic power system,” IEEE Trans. Power Electron., vol.
Newcastle upon Tyne, U.K. He is also with the School
25, no. 4, pp. 992-1000, Apr. 2010.
of Automation Engineering, University of Electronic
[38] P. Guerriero, V. d’Alessandro, L. Petrazzuoli, G. Vallone, and S. Daliento,
Science and Technology of China, Chengdu, China. He
“Effective real-time performance monitoring and diagnostics of individual
has coordinated several research projects from the Engineering and Physical
panels in PV plants,” the 4th International Conference on Clear Electrical
Sciences Research Council, Royal Academy of Engineering and FP7. He also
Power (ICCEP), Alghero, Italy, pp. 14-19. 2013.
has collaborative links with leading industrial companies, such as Airbus, Rolls
[39] P. Guerriero, G. Vallone, M. Primato, F. Di Napoli, L. Di Nardo, V.
Royce, BP, nPower, and TWI.
d’Alessandro, S. Daliento, “A wireless sensor network for the monitoring
of large PV plants,” the International Symposium on Power Electronics,
Electrical Drives, Automation and Motion (SPEEDAM), Ischia, Italy, pp.
Stephen J. Finney obtained the M.Eng in electrical
960-965, 2014.
and electronic engineering from Loughborough
University of Technology in 1988; and the Ph.D.
Yihua Hu (M’13) received the B.Sc. in electrical
degree from Heriot-Watt University, Edinburgh, U.K.,
motor drives in 2003, and the Ph.D. in power
in 1995. He worked for the Electricity Council
electronics and drives in 2011, both from China
Research Centre before joining the Power Electronics
University of Mining and Technology, Jiangsu, China.
Research Group at Heriot-Watt University in 1990.
Between 2011 and 2013, he was with the College of
From 1994 to 2005 he was a member of academic staff
Electrical Engineering, Zhejiang University as a
at Heriot-Watt University. Since 2005 he is with the
Postdoctoral Fellow. Between November 2012 and
Institute of Energy and Environment, University of
February 2013, he was an academic Visiting Scholar
Strathclyde, Glasgow, U.K., where he is currently a
with the School of Electrical and Electronic
Professor in power electronic systems. His research interests include the power
Engineering, Newcastle University, Newcastle upon
electronics for high power applications, power transmission and distribution.
Tyne, UK. He is currently a Research Associate with the Department of
Electronic & Electrical Engineering, University of Strathclyde, Glasgow, U.K.
His research interests include PV systems, DC-DC/DC-AC converters, and
electrical motor drives.

9
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
James L. Kirtley Jr. (LF’91) is a Professor of
electrical engineering at the Massachusetts Institute of
Technology, Cambridge, MA, U.S.A. He was with
General Electric, Large Steam Turbine Generator
Department as an Electrical Engineer, and with Satcon
Technology Corporation as Vice President and
General Manager of the Tech Center, and Chief
Scientist and Director.
Dr. Kirtley attended MIT as an undergraduate and
received the Ph.D. from MIT in 1971. He was
Gastdozent at the Swiss Federal Institute of Technology, Switzerland. He
served as Editor in Chief of the IEEE TRANSACTIONS ON ENERGY CONVERSION
from 1998 to 2006 and continues to serve as Editor for that journal, and as a
member of the Editorial Board of the Electric Power Components & Systems
Journal. He was awarded the IEEE Third Millennium medal in 2000, and the
Nikola Tesla prize in 2002. He was elected to the United States National
Academy of Engineering in 2007. His research interests are electric machinery
and electric power systems.

10