Online I-V Tracer for per string monitoring andmaintenance of PV panels

Ashish V. JoglekarRobert Bosch Center for Cyber Physical Systems

Indian Institute of ScienceBangalore, Indiaashishj @iisc.ac.in

Abstract-Large scale solar photovoltaic plants need to bemonitored regularly for determining location and type of fault,cleaning schedules, operating point and performance. We proposea circuit that sits at the junction box level to provide 'on demand','online' I-V traces for series connected panels. This sparse sensingapproach works on a per string basis and is therefore economical.The proposed online I-V tracer topology does not require theplant to be brought offline to obtain the I-V trace. The load'spower requirement is met during the trace. The shape of the I-Vtrace helps determine the type of fault and localizes the faultto a specific string. The proposed solution has been tested ina practical field deployment. An analytics engine is also beingdeveloped to use the recorded I-V curve to provide optimalcleaning schedules, fault diagnosis and maintenance alerts.

Index Terms-solar power generation; solar panels; conditionmonitoring; fault diagnosis; I-V curves; I-V tracers; circuittopology

I. INTRODU CTION

Large scale PV installations need to be monitored andmaintained to ensure uninterrupted operation at their optimumcapacity [I].

There are many possible failure modes in PV installations[2] [3]. Partial and/or full shadowing due to dust accumulation(soiling) has been reported to reduce plant output by 34% un-der specific conditions in [4]. Consequently, periodic cleaningof the panels becomes a necessity . Hot spot generation (due tocell mismatch or shading) should be detected early to preventpermanent damage to the cell [3]. Faults in cabling like seriesinterconnect faults (due to for example corrosion) and shuntleakage faults (due to metal migration) also result in reducedplant output. Bypass diode thermal failure (due to overheatingor under sizing of diodes) can lead to reduced power outputand faster degradation of cells. Deterioration in anti-reflectivecoating can also bring down plant output [3].

Fault detection techniques using instrumented referencepanels , infrared cameras on drone s [5], statistical comparisonof multiple string outputs, surface temperature, output current,output voltage and irradiance measurements and I-V curvemeasurements have been reported in literature. Among thesetechniques, I-V curves can be used to identify a large rangeof the aforementioned faults. The shape of the I-V trace canbe mapped to a specific type of fault [6]. There are manytechniques to obtain an I-V trace from a PV installation [7].

Balachandra HegdeRobert Bosch Center for Cyber Physical Systems

Indian Institute of ScienceBangalore , India

b.m.hegde @rbccps.org

The common techniques use a resistive load or a capacitiveload [8] [9] or an electronic load [10] or a DC-DC converter[II] to obtain the I-V trace.

During commissioning and yearly maintenance of a plant ,I-V tracers are commonly used to determine the health ofthe panels/strings [12]. During this operation, the plant isbrought offline. An impedance sweep then reveals the I-V trace . Commercially available/ portable I-V tracers arevery expensive and require the load to be disconnected fromthe panels during the I-V trace [II]. The challenge lies indesigning a low cost , online I-V tracer that can record the I-V curve on demand from an online PV installation withoutinterrupting the power flow to the load. This approach of I-V curve based plant monitoring can be effectively used forpre-emptive plant maintenance thus reducing downtime andmaximizing output.

Few online I-V tracers have been previously reported inliterature. These IV tracing techniques are limited in theirability to provide accurate, on demand IV traces. A SEPICbased online I-V tracer for single PV module was proposed in[II] . This DC-DC converter based approach provides limitedaccuracy due to inherent current ripple and DCM operationresults in a trace time of lOOmsec. Only simulation resultshave been provided. Reference [13] describes a capacitor basedI-V tracer which also works in conjunction with a MaximumPower Point (MPP) Tracker. Their approach assumes that loadsare equipped with storage devices which continue to provideuninterrupted power when panels are isolated from the loadsduring the trace . A bleed resistor discharges the capacitorbefore the next I-V trace. In [14], a model is proposed thatestimates a PV panel's I-V trace using the panel 's instanta-neous DC voltage , DC current and environmental data . Theapproach requires instrumentation of every panel. The modelonly estimates three points (i sG, Va G and NI P P ) on the I-Vtrace . In [15], few I, V data points are recorded using an MPPTconverter to approximate the 1-V trace. These approximated IVtraces are limited in their ability to diagnose various failuremodes. Fault diagnosis that relies on accurate sensing of i s c ,

determination of horizontal or vertical leg slope and roundnessof knee cannot use an approximated trace .

We propose a capacitor as load based , fast, low cost ,retrofittable, online I-V tracer that sits at the junction box level

978-1-5090-6684-1/18/$31.00©2018 IEEE 1890

: /r

F/

1/:

i : :

B±'ru." !~ " ,........ , ••p .. ' "'_. e, ~ 500;!t .. ...,. .. 1.. . .

0. 120.110.100.09

t (sec)0.080.07

175

150

~ 125

G 100;".

~ 75

50

25

o87654321o

-10.06

RS2

~~

".

S S2 ._0.S~

Fig. 1. Proposed Online I-V tracer topology.

Fig. 3. Panel voltage Vpv and panel current i ev . These waveforms arecaptured during state A or C. Also shown are phases in the charging cycle ofthe capac itor (while performing the I-V trace).

to provide I-V traces on a per string basis.Our unique topology meets power requirement of the load

while record ing the I-V trace. We utilize the energy stored inthe capacitor during the trace to provide uninterrupted powerto the load. The main advantage of our IV tracer topologyis its ability to provide an on demand, 1500 point, accurateIV trace while the plant is online. The ease of deployment atjunction box level is another plus. However, a fault can onlybe resolved to a PV string when our IV tracer is placed atjunction box level. A prototype of our online I-V tracer hasbeen deployed in a practical field installation.

We are also working on an analytics engine that uses therecorded I-V trace to optimize plant cleaning schedules (thusreducing usage of water required to clean panels), reduce

v ~

~ ,

t ~ .'

v ~

repair costs, improve plant output and efficiency.

II . PRO POSED O NLI NE 1- V TR ACER T OP O LOGY

A. Circuit Description

Fig. I shows the proposed online I-V tracer topology con-sisting primarily of four switches 8 1 to 8 4 and two capacitorsC1 and C2 . Fig. 4 shows the practical implementation of theproposed I-V tracer. We have chosen 470fJP capacitors. Vpvis the PV panel side DC voltage bus. ZL is any downstreamload. RBI, 8 BI and RB 2 , 8 B2 are bleed discharge paths toremove any residual charge on the capacitors C1 and C2 • Thesedischarge paths are activated during specific operational statesof the tracer. Switches 8 1 to 8 4 are implemented as backto back N channel enhancement mode MOSFETs result ingin a true unidirectional switch. The high side gate driverequirements are met with a separate isolated DC bus andan opto-coupler based gate drive circuit. Parameters measuredare capaci tor currents i C 1, i C 2, panel voltage Vpv and loadvoltage VL. Currents i C1 , i C2 are measured using a currentsense resistor, an analog front end for signa l conditioning, anda SAR ADC. A microcontroll er is programmed to generate therequired gate dr ive signals and record measured parameters.The entire circuit is powered from the PV panels. During thetrace panel voltage drops to zero . The DC power bus has

Fig. 2. State Transition Diagram for proposed online I-V tracer. Fig . 4. Prototype Online I-V tracer.

1891

Fig. 5. I-V trace recorded by online tracer from a test deploy ment.

eno ugh buffer capac itance to keep the circuits powered duri ngthe trace . Whi le one arm performs an I-V trace the other armmeets the load's power req uirement. The work ing princip le isexp lained in the state transition diagram in Fig . 2

0.3000.2750.2500.175 0.200 0.225

t (sec)0.1500.125

"'-

I I I I I I

7I

-III

175

150

~ 125

G 100;"

~ 75

50

25

o175

150

125

~ 100

.:;: 75

50

25

o0.100

Fig. 6. Panel voltage Vpv and load voltage VL during I-V trace. Thesewaveforms are captured during state A or C

~ W 100 IW I~ I ~ IW

Vpv (V )4020

I I I I I I I ~ I~

1,2

oo

8

6

~ W 100 IW I~ I~ IW

VPv(V)4020

O+----+----+- -i----+-----+-------T- -;-----"'_T_- ----jo

Fig. 7. I-V trace recorded by online tracer from a test deployment underpartial shading conditions .

the I-V trace, whi le capaci tor C2 meets power requirement ofthe load. The cycle repeats again starting with state B.

Intermediary states that ensure reliable state transitions arenot shown in the state transition diagram of Fig . 2.

Va Gt s can = C -.- (I)

2S G

The t s can time for our proposed circuit is < 20msec. Asper Ref. [16] this scan time results in < 0.5% error in I-Vmeasurement (for poly crystalline Si pane ls).

Scan time is seen to be a function of irradiance. The I-V tracing capacitor will charge at a faster rate at higherirradia nce. As it takes longer to comp lete a trace at lowerirrad iance, an irradia nce threshold can be set. No trace will

C. I- V Trace

We get two I-V traces per cycle at state A and state C. Onesuch I-V trace obtained from the field is shown in Fig. 5.

The time taken for comp leting the trace is approximatelygiven by (I) [9].

B. Operational States

Consider the state transition diagram shown in Fig. 2 Letus start with state B. In this state switches SI and S2 are ON.Thus, the panels direct ly con nect and supply power to theload ZL. The voltage across capacitor C1 is equal to panelvoltage Vpv in this state. In this state switches S3 and S4are OFF. Swi tch S B2 is ON to remove any residual charge incapacitor C2 via bleed resistor RB 2 . Thus, at the end of thisstate, capacitor C2 will be discharged completely.

In the next state C, switc h SI turns OFF whi le switch S3turns ON. As panels are disconnected from the load ZL , thepower requirement of the load is now met solely by chargedcapaci tor C1 . Capac itor C2 , which was discharged in previo usstate, is now the new load across the pane l. A dischargedcapaci tor appears like a short across the pane l. Capacitor'scharging current and pane l voltage waveforms are shown inFig. 3. In phase I, the pane l voltage drops to zero and rampsup linearly as the pane l current of value rv i S G charges thecapac itor. In phase 2 pane l current quick ly falls to zero asthe capacitor voltage tends to the string's open circuit voltageVaG. The trace time is seen to be < 20m sec. A chargedcapaci tor C2 is equivalent to an open circuit across the PVstring. The impedance across the pane l is effective ly sweptfrom short circuit to open circuit condi tion to revea l thestring's I-V trace. The panel voltage and current are recordedduring the capacitor's charging cycle. At the end of this state ,capacitor C2 would have charged to Va G.

In the next sta te D, both switches S3 and S4 are ONconnecti ng the pane l directly to the load ZL. The voltageacross capacitor C2 is pane l voltage Vpv in this state. SwitchesSI and S2 are off in this state. The residual charge on capac itorC1 is removed via bleed resistor RBI as switc h SBI is ON inthis state. Thus, at the end of this state, capacitor C1 will bedischarged comp letely.

In the next state A, switch SI turns ON and S3 turns OFF.In this state, capacitor C1 appears across the pane l to provide

1892

------------- I

~I<. < <I

ipv I

I I- •• -A I AI

I

PV String --r --r I

I I

Vpv i I

i VL LoadIL ____________ _ _ I

Online IV Tracer

1Junction Box

Fig. 8. Field Deployment of proposed online I-V tracer at jun ction box level.

monitored during the trace to ensure that it does not fallbelow a set threshold. If the load voltage hits the predefinedthreshold value, I-V tracing is terminated prematurely as afailsafe feature. Usually, the I-V tracing capacitor charge s ata faster rate than the discharging capacitor in the other arm ascharging current for the I-V tracing capacitor is rv i s c , whilethe other capacitor discharges at a current typically < i sc . Theplot in Fig. 6 shows the panel voltage Vpv during state A (orC) and load voltage VL . It is seen that load voltage VL doesnot fall to zero when the panel is disconnected from the load.As one arm traces the I-V curve, the capacitor in the otherarm maint ains the DC bus of the load. Energy stored in thecapacitor during the I-V trace is thus fed to the load in thefollowing state.

III. FIELD D EPLOYM ENT

Fig. 8 shows the position where the online I-V tracer isdeployed in a practical installa tion. Our online I-V tracer isinterposed between the SPY string and the load. If the SPYstring's voltage exceeds the tracer 's voltage rating , the tracercan be placed at an intermediate position in the string. Thetracer will then provide an I-V trace for a subset of the seriesconnected panels in the string (panels upstream).

Our online I-V tracer sits at the junction box level. Thetracer is rated for 200V/ IOA . Each SPY string comprises ofeight 21.5V, 5.1A panels. Two such strings are connected inparallel to result in a Vac = 172V and an i sc = lO.2A. Thecaptured trace from this installation is shown in Fig. 5. Weare currently using a resistive bank of minimum value lOOnas the load for the test.

IV. FAULT DIAGNOSIS

We are working on an analytics engine that takes in therecorded I-V trace and the current, irradiance and temperature

value. The trace is compared with the expected ideal traceunder the measured irradiance and temperature conditions. Theengine will output probability that a given fault has occurred,provide cleaning schedules and maintenance alerts.

An increase in horizontal leg slope typically indicates ashunt leakage fault, a reduced vertical leg slope indicate s aseries interconnect fault , reduced i sc for given irradiance andtemperature conditions is an indicator of shadowing of thepanel due to dust accumulation. Partial shadowing can lead tomultiple distinct steps in the I-V plot as current takes the pathof bypass diodes . Fig. 7 shows one such plot captured underpartia l shadowing conditions using our online I-V tracer.

Once the failure is localized to a string using the online I-Vtrace, techniques like time domain reflectometry [17] can beused to detect the exact panel where fault lies.

V. ADDITIONAL F UN CTIONALITY

The same circuit can also provide data about plant effi-ciency, Maximum Power Point Tracking etc. The I-V plotreveals the maximum power point for the panel with completeinformation of local and global maxima. The circuit canalso play an additional role of a battery charge controlleras it already monitors the voltage on the load side and hasunidirectional series switches to regulate power flow.

VI. CONCL USION

This paper describes an I-V tracer circuit that provides'on demand' I-V traces on a per string basis from an onlinePV installation. The circuit topology allows energy stored inthe capacitor during the trace to be delivered to the load inthe following state. The load does not experience any powerinterruption while the panels are isolated from the load duringthe trace. The circuit is retrofittable at junction box level.

1893

The prototype tracer will be installed in a larger PV plantand an analytics engine will be developed to provide mainte-nance alerts and cleaning schedules based on the shape of theI-V trace.

A CK NOWL EDGMENT

The authors thank Prof. Bharadwaj Amrutu r, Chairman,Robert Bosch Center for Cyber Physical Systems, IISc forsupporting this project.

R EFER ENCES

[I] Hatti, Mustapha. "Operation and Maintenance Methods in Solar PowerPlants," in Use, Operation and Maintenance of Renewable EnergySystems, pp. 6 1-93. Springer, Cham, 20 14.

[2] Nilsson, David. "Fault detection in photovoltaic systems," Dissertation20 14.

[3] Deline, Chris. "Guide To Interpreting IV Curve Measurements of PVArrays." Solmetric Application Note 20 I I.

[4] Al-Hasan, Ahmad Y., and Adel A. Ghoneim. "A new correlation betweenphotovoltaic panel's efficiency and amount of sand dust accumulatedon their surface." International Journal of Sustainable Energy 24, no. 4(2005): 187-197.

[5] Aghaei, Mohammadreza, Francesco Grimacc ia, Carlo A. Gonano, andSonia Leva. "Innovative automated control system for PV fields inspec-tion and remote control." IEEE Transactions on Industrial Electronics62, no. I I (20 15): 7287-7296.

[6] Meyer, Edson L., and E. Ernest Van Dyk. "Assessing the reliabilityand degradation of photovoltaic module performance parameters." IEEETransactions on reliability 53, no. I (2004) : 83-92.

[7] Duran, E., M. Piliougine, M. Sidrach-de-Cardona, J. Galan, and J. M.Anduja r. "Different methods to obtain the I-V curve of PV modules: Areview." In Photovoltaic Specialists Conference, 2008. PVSC 08. 33rdIEEE, pp. 1-6. IEEE, 2008.

[8] Erkaya, Yunus, Paul Moses , and Sylvain Marsillac. "On-site charac-terization of PV modules using a portable, MOSFET-based capacitiveload." In Photovoltaic Specialists Conference (PVSC), 20 16 IEEE 43rd,pp. 3 119-3122. IEEE, 20 16.

[9] Warner, T. H., and C. H. Cox III. "A high power current-voltage curvetracer employing a capacitive load." Solar Cells 7, no. 1-2 (1982): 175-181.

[10] Papageorgas, P., D. Piromalis, T. Valavanis, S. Kambasis, T. lliopoulou,and G. Vokas. "A low-cost and fast PV IV curve tracer based on an opensource platform with M2M communication capabilities for preventivemonitoring." Energy Procedia 74 (2015): 423-4 38.

[II] Riley, C. w., and L. M. Tolbert. "An online autonomous I- V tracerfor PV monitoring applications." In Power and Energy Society GeneralMeeting, 20 15 IEEE, pp. 1-5. IEEE, 2015 .

[12] IEC 62446- 1 Standard "Photovoltaic (PV) systems - Requirementsfor testing, documentation and maintenance - Part I: Grid connectedsystems - Documentation, commissioning tests and inspect ion"

[13] Spertino, Filippo, Jawad Ahmad, Alessandro Ciocia, Paolo Di Leo, AliF. Murtaza, and Marcello Chiaberge. "Capacitor charging method forI-V curve tracer and MPPT in photovoltaic systems." Solar Energy 119(20 15): 46 1-473.

[14] Molina-Garcia, Angel, Jose Carlos Campelo, Sara Blanc, Juan JoseSerrano, Tania Garcia-Sanchez, and Marfa C. Bueso. "A decentralizedwireless solution to monitor and diagnose pv solar module performancebased on symmetrized-shifted gompe rtz functions." Sensors 15, no. 8(20 15): 18459-18479.

[15] Manjunath, Matam, B. Venugopal Reddy, Y. Zhao, and Brad Lehman."On-line health monitoring of PV plants." In Energy ConversionCongress and Exposition (ECCE) , 2017 IEEE, pp. 406 1-4068. IEEE,20 17.

[16] Herman, Marie, Marko Jankovec, and Marko Topic. "Optimal IV curvescan time of solar cells and modules in light of irradiance level."International Journal of Photoenergy 20 12 (20 12)

[17] Takashima, Taku mi, Junji Yamaguchi, Kenji Otani, Kazuhiko Kato, andMasayos hi Ishida. "Experimental studies of failure detection methodsin PV module strings." In Photovoltaic Energy Conversion, ConferenceRecord of the 2006 IEEE 4th World Conference on, vol. 2, pp. 2227-2230. IEEE, 2006

1894