Diagnose errors on photovoltaic modules.

Đăng ngày | No Comments


Fault diagnosis and condition monitoring are important to increase the efficiency and reliability of photovoltaic modules. This paper reviews the challenges and limitations associated with fault diagnosis of solar modules. A thorough analysis of various faults responsible for failure of solar modules has been discussed. After reviewing relevant work, a monitoring tool is designed using thermography and artificial intelligent systems that allows the detection of various types of faults in PV modules and at the same time the designed tool aims to filter the nonsignificant anomalies. A neural network (NN) classifier is applied to the transfer characteristics (I-V data) of the faulty PV module for the diagnosis which adapts multilayer perceptron (MLP) networks to identify the type and location of occurring faults. The Discrete wavelet transform (DWT) based signal processing technique is utilized in the feature extraction process to reduce the NN input size. The developed detection algorithm is adapted for 24/7 automated surveillance. For a given fault condition, the average fault detection time is observed to be <9 seconds, which is lower than the previous work done. The developed algorithm achieved 100% accuracy when tested on a predetermined fault data set.


In recent years, there has been an exponential growth in photovoltaics across the world. This brings many problems associated with the quality of the systems due to several factors. The most crucial component in PV systems is the photovoltaic module whose diagnostics can be sometimes relatively difficult. The main problem is that all the modules look very similar although their quality is totally different. Defects on the modules are usually not visible by the naked eyes and also their causes should be found by the special methods. As a result, there are many installations that necessarily requires the regular monitoring. The Present-day statistics depicts the rate of degradation for power ratings in crystalline silicon photovoltaic modules by 0.8%/y.1 By observing the challenges involved with the PV modules and their operation, work can be carried toward the improvement in reliability and the service life of PV modules. In general terms, the faults observed in any component are classified under three categories namely, Early failures, Intrinsic or Random failure, and Deterioration.23 Early failures are observed during the initial stages of installation and operation of PV module. Any defect or flaw in the PV module results in quick degradation effecting the service life of the module. Of different fault reasons, the optical failures, and the Junction box failures are commonly observed. The extrinsic failures observed in PV systems are also known as midlife failures.4 It was predicted that 2% of the PV modules does not comply with the warranty of manufacturer after a certain span of operation. Various studies depicted that the interconnection defects and module Glass breakage failures are the major causes for extrinsic failures. David DeGraaff5 presented the relative failure rates of various PV system components. The research stated that the life of the PV module is concealed whenever, a safety problem or power drop occurs. The power drop of a PV module is typically defined between 80% and 70% of the systems initial power rating. Freire et al6 reviewed the degradation modes by investigating on some PV modules with 10 years of activity. The most common failures in a PV module are laminate discoloration, isolation of cell parts due to cracks, and delamination. It was observed that these failures resulted in about 10% mean power loss.7 During the conventional manufacturing process of PV modules, the inability of manufacturers to check the cells for cracks and metallization had helped in obtaining knowledge about long term degradation mechanism for today’s PV modules. The further sections of the paper deal’s with different module failures, conventional fault detection methods, their drawbacks, developing reliable fault detection, and classification techniques.


Failure in general terms is explained as an effect that degrades the ability of a system and cannot be inverted by normal operation. The same scenario applies here with the PV systems. Any abnormal operation that occurs under normal operation of a PV module is relevant for the warranty and is termed as PV module or system failure. Manufacturing defects are considered as the main reason for instabilities in performance of some modules. Some of the examples of manufacturing defects can be observed in mono and multi crystalline solar cells in the form of striation rings and moderate crystal defects, respectively. In crystalline silicon modules, for achieving power rating boron oxygen compound is used.8 Any mismatch in the proportion of compound, results in light induced power degradation and is termed as manufacturer defect for PV failure. Due to failure of module technology, the amorphous silicon modules are prone to light induced degradation accounting for a power loss of 10%-30% during the initial stages of installation.9 This degradation can be recovered to some extent with the application of thermal annealing,10 but it is applicable only during the hot summers and any seasonal variation will result in degraded performance of the module.

2.1 Faults due to external causes

Apart from manufacturing defects, there were many others which are characteristically difficult to classify either as manufacture failures or even for other reasons. Some of the module failure due to external causes were, Transportation Failure, Clamping, Cable failure, connector failure, and Lightning. The effect of transportation on the PV modules were cited by.11 It was observed that the breakage of glass cover and lamination damage in some modules are due to shocks and vibrations during transportation. It is clear that this failure does not accommodate with manufacturing defect and is one of the major external causes of module failure. Most of the transportation failures can be identified neither visually nor by observing power ratings. Lock-in thermography image or an electroluminescence image can detect such damages. Among the installation issues associated with modules, clamping is the most often failure which results in glass breakage mostly for frameless PV modules. Figure 1 depicts the effects of poor clamping in PV modules. Yixian Lee et al12 conducted finite element analysis to observe the stresses faced by PV modules especially during installation stage. Various observations depicted that the sharp-edged clamp design, narrow clamps, improper positioning, and excessive tightening of screws on clamps of module causes stress on PV modules resulting in breakage. The effects of Glass breakage results in electrical safety issues and performance loss during long time operation as corrosion may occur due to moisture penetration through the cracks. The cracks developed also lead to hot spots, resulting in module overheating.1314

Details are in the caption following the image
Figure 1: PV module that broke due to poor clamp design

Cables and connecters are associated with PV systems for providing electrical connections to solar modules and other components of PV system including Inverter. Connectors are very important elements and play a major role in safety and reliable power generation and transmission. Of different connectors, Low-voltage DC connectors were widely discussed due to their usage in electric vehicles and in PV systems.15 Claudio Ferrara et al3 mentioned about connectors and different metals associated with them, which can show defects and cause of corrosion when exposed to atmospheric humidity alone or in combination with gases. Considering all the effects and consequences, the constraints due to cables or connectors used in a PV system were not considered as manufacturing defects. Conventionally connector failures are observed in the cases of improper cable selection or inaccurate connections between PV modules and components associated with them. These failures sometimes cause total power loss in the string and can lead to electric arcs and fires. Zaini et al16 studied the effects of lighting on a grid integrated PV system with an assumption that the lightning impulse current strikes the PV system at two different points as shown in Figure 2. The effect of lightning strike on the DC side ie at the module is observed in the form of a defective bypass diode. This effect tends to be an external cause resulting in a subsequent safety failure. Such external causes mostly result in open-circuit bypass diodes or module failure due to direct exposition to lightning. Different faults observed in a module due to both manufacturing defects and external causes were discussed in the further sections.

Details are in the caption following the image
Figure 2: Lightning impulse current striking at different points

2.2 General faults in all PV modules

In general, most anticipated failure modes and degradation mechanisms were related to glass breakage, junction box failure, interconnection faults, and delamination. Juris Kalejs17 raised concern regarding robustness and durability of junction boxes. The research stated that improper design or improperly disclosed junction boxes ingress moisture which causes corrosion to connections in the junction box. This causes wiring failure which leads to internal arcing. Uichi Itoh et al18 depicted the potential risks associated due to soldering failures in junction boxes. The observations depicted two different faults, ie silver (Ag) leaching and solder joint fatigue. When solder joints come in contact with the AG electrodes of solar cell, they get dissolved with the solder electrodes (Tin–Lead (Pb–Sn)) and is observed as Ag3Sn compound.18 This Ag leaching effect develops crack in the soldered interface due to thermal expansion resulting in connection break down. Delamination effect occurs due to adhesion contamination or due to environmental factors as shown in Figure 5, which ingress humidity, moisture, and corrosion into the laminates of PV module. The lamination failures result in optical reflection which results in subsequent power loss from the modules. Kleiss et al19 reported the quality and reliability of the PV modules and stated that 90% of modules are prone to delamination. Zhu et al20 observed the causes of delamination and stated the adhesion requirements that are to be met by the manufacturers. Issues with variation in lamination temperatures and adhesive materials were disclosed to be the major concerns for delamination. Tracy et al21 proposed a novel test procedure apart from the conventional peel test to detect delamination. The delamination phenomenon can also be observed in thin film modules where the transparent conductive oxide delaminates from the glass layer.22 In general, delamination can be relatively observed, with the help of reflectometer. Visually, delamination’s can be identified by using lock-in thermography, pulse thermography, X-Ray thermography, and an ultrasonic scanner.2324 The Photovoltaic (PV) back-sheets provide safe operation during high voltages and protect electronic components from being exposed to aggressive field environments.25 Back-sheets are manufactured with different materials like polymers, glass, and metal foils. Mostly back-sheets are manufactured using highly stable laminate structures and UV resistant polymers. The materials were chosen depending on the mechanical strength required, cost, and electrical isolation. Singh et al26 mentioned about using rear glass with back-sheet as the structure offer more power. There were some serious issues associated with this structure as any improper mounting or mechanical stress will lead to breakage of the glass. Irrespective of the above-mentioned problems, glass/back-sheet structure can offer 2%-3% higher power when compared with standard back sheet modules. Tang et al27 presented a double glass PV module which can withstand various environmental conditions due to the “0” moisture permeable rate and exhibits long-term stability and reliability. A composite Ageing test was conducted for performance analysis which proved to be better than the standard PV modules. Polymeric laminates are most commonly observed back-sheet construction materials. They have multiple layers which causes the effect of delamination of interfaces due to high physical and chemical stress. The only advantage that can be seen in the failure of polymer laminates is that, when delamination occurs, it will not create an immediate safety issue. Back-sheet delamination near a junction box tends to be a major issue as it results in an unconstrained junction box which leads to positioning of mechanical stress on the live components and breaking them. This breakage further results in connection failure at bypass diode which is further complicated by formation of an unmitigated arc at full system voltage.

2.3 Review of failures found in silicon wafer-based PV modules

Crystalline silicon wafer-based PV modules share a dominant market in the world of PV modules due to their wide spread applications.28 These modules hold 95% market share as of 201729 and is the most widely used solar cell type. Despite of their widespread applications and usage, these modules are prone to very common failures like potential and light induced degradation and snail tracks. Discoloration of Ethylene Vinyl Acetate (EVA) encapsulants were initially observed at Carrizo plains installation sites in California around early 1990s and was observed to be a major problem.1330 Chianese et al31 discussed about ASI 16-2300 single crystalline silicon photovoltaic modules which use poly vinyl butyral encapsulant and tedlar/aluminum/tedlar back sheet. It was observed that the developed structures require a robust electrical insulation layer between the cell and the foil, which raises numerous safety concerns. Apart from the electrical insulation, metal foils also act as high voltage capacitors and any disturbance towards the electrical isolation of foil surface will charge the foil at system voltage. Pern32 investigated the various factors that affect degradation rate of EVA polymer encapsulants using noninvasive analysis. The analysis observed the field degraded, yellow to dark brown EVA to understand the chemical and physical damages. EVA is consistently drafted with additives like cross linking agent, antioxidants, UV absorber, hindered amine light stabilizers and adhesive elements. Peike et al33 observed that these additives formulate for the root cause for discoloration by originating chromophore and lumophores. The discoloration results were noticed in different patterns and were examined for complexity due to oxygen diffusion and acetic acid from additive reactions. The origin of chromophores results in a transparent EVA ring around the edges of a wafer-based cell. In some scenarios of EVA discoloration, it is observed that a single cell in the modules is gloomier than other cells. This usually defines that the temperature sensitivity of the particular cell is greater than the adjacent cells due to low photocurrent or because of the cell being placed above the junction box.34 The EVA discoloration in severe cases corresponds to EVA embrittlement, and concomitant corrosion due to dissolved oxygen.35 Even though most of the module failures cause an outright failure, discoloration and delamination does not lead to a failure, but degrade the functionality at a very slow degradation rate of ~0.5%/a.3637 During severe discoloration a total loss of ~10% is observed which implies that EVA discoloration is absurd for a complete failure of silicon modules. Dhimish et al38 reviewed various types of cell cracks and their effects and proposed a statistical analysis-based approach to identify them. The observations revealed about the various types of cracks namely Multi Directional Cracks, Diagonal cracks, and cracks which are parallel and perpendicular to bus bars. In case of cracks in a cell, the element might not be totally disconnected from the cell but develops a resistance between the number of cycles present in the deformed module and the cell elements. Kajari-schroder et al39 analyzed the criticality of cracks and stated that, cracks in PV modules lead to low stability in output power under artificial aging. Experimental analysis was conducted on 667 cracked cells in 27 PV modules and the results depicted that 50% of crack orientations were parallel to busbar cracks which are considered for high criticality. The variation in manufacturing process of solar modules also leads to cell cracks specifically during the stringing process.40 Issues related to transportation and installation of PV modules were also depicted as a major contribution towards cell cracks. Meyer et al41 investigated PV modules with snail tracks formed due to outdoor exposure of defected modules for several months. A detailed microscopic imaging of discoloration helps in identifying the failure. It was observed that snail trails were mostly located at cell edges or near micro cracks. It is not necessary that all the micro cracks will be developed to snail trails, but whenever a snail trail is observed in a module or cell, a micro crack is found at the same position. The optical impression of the trail is due to brownish discoloration of the grid finger position which is further imprinted to the EVA foil. The original process of discoloration due to snail tracks is not well defined in the literature. The optical impression of the snail tracks varies in different modules disturbing the cell fragments and cell edges. The formation of snail tracks in a module is termed as an electromechanical degradation process but was never deemed to be a direct cause for power loss. Fairbrother et al42 depicted that in a PV array 3% of the PV modules were subjected to burn marks at two different positions occupying 5% area of the back sheet. In the case of Back sheets, two types of burn marks are observed on the basis of temperature. One type of burn marks observed has high rupturing capacity due to high temperature, while the other types are of low rupturing capacity with lower temperatures. When burn marks were closely observed they depict physical damage through scratches, scrapes and tears. Friere et al6 stated that the relative failure rates of burn marks on a cell is ∼10%. Mohamed et al43 observed that the average annual degradation of PV modules power is 1.5% caused due to failures like burn marks, cell cracks, and delamination. It was depicted that most of PV module failures doesn’t exhibit visible burn marks but result in severe power loss.44 Pingel et al45 mentioned that Potential Induced Degradation (PID) is caused due to polarization effect or due to chemical corrosion. The level of degradation depends on the polarity and level of potential difference between PV cell and the Ground. PID is also responsible for durability issues in modules. In general, PID is observed when high voltages force sodium ions to spread out from the glass through encapsulate and accumulate on cells surface. This results in surface recombination, fill factor drop and increased local shunting.46 There are two types of PID, nonreversible and the reversible PID. The nonreversible PID occurs due to electrochemical reactions resulting in electro corrosion of transparent conducting oxide. The reversible PID also known as surface polarization accumulates positive charge on a PV cell which results in leakage current. Depending on the grounding configuration of a PV array the amount of leakage current is defined. This degrade the generation capability of the solar cell. Typical circumstances for the occurrence of PID is observed at different levels like environmental factors, module factors, system factors, and cell level. These factors are generally dependent on the temperature, humidity, system voltage, type of material used, and the refractive index of the cell. Cristaldi et al4748 stated that electrochemical corrosion of the string interconnects of a cell due to encapsulant results in the degradation of PV module. This phenomenon leads to high series resistance and low parallel resistance of the PV modules.49 The PV cells are generally equipped with front and rear contacts connected through bus stripes for delivering current to the external circuit. Any failure in the string ribbons result in thermal expansion—contraction and mechanical stress resulting in output power loss.5051 Hermanan52 and Buero53 depicted the influence of ambient temperature and temperature difference caused by flowing current on diode temperature. From the real operation, it is known that the temperature of the module is approximately about 20°C higher than the ambient temperature. As the diodes are present inside the junction box, there is an immediate vicinity of the module surface, maintaining very similar temperatures. It means that in idle state, the diode will have the ambient temperature + 20°C. In clear summer days about 55°C and in winter about 20°C.54 When the ambient temperature is low, the current flowing through the diode will be lower because of the higher value of threshold voltage. Also, the heating by the flowing current will be then limited. When the ambient temperature is high, the threshold voltage value will decrease and it will cause higher flowing currents and another additional bypass diode heating.

2.4 Faults in thin-film modules

In order to save materials (and thus costs), thin-film PV modules have been developed. These modules, against conventional crystalline modules, have the lower conversion efficiency. But this is offsets by lower prices (production is less materially and usually less technologically demanding) and improved properties at low irradiance levels. From the viewpoint of the material and manufacturing process, the modules are classified as, CuInSe2 (CIS), Cu (In, Ga) Se2 (CIGS) a CuGaSe2 (CGS) modules, CdTe modules, amorphous and micromorphous silicon modules, and many other thin-film cells like multijunction cells, cells using nanostructures, and organic cells. The design process for all the above-mentioned modules varies as they deal with multiple compounds and compositions. But irrespective of the design procedure the cause and effect of any fault on a given module has a potential effect which degrades the module. Some potential cause and effects that were observed for thin film modules are glued connectors micro arcs, hot spots shunts,55 front glass breakage, and back contact degradation.5657 Recently, first solar published an introduction to the subject58 and interesting degradation kinetics. Depending on climate and system interconnection factors, one may expect an initial degradation of 4%-7%, over the first 1 to 3 years. Various faults that can be seen without the help of any instrument are depicted in Table 1:

Table 1. Visual faults associated with a PV module2341
Fault type Power loss Safety issue Visual fault
Shorting of module wires and diodes <3% of power loss Fire failure may be caused image
Laminated cell fragment <3% of power loss Fire failure, electric shock, and physical danger image
Cell cracks damaging 10% of cell area Degradation of power loss which saturates over time No effect on safety image
Bubbles or delamination Degradation of power loss in steps over time Electric shock resulting in major safety problem image
Burn marks on back sheet Degradation of power loss in steps over time Failure may cause fire, electric shock, and physical danger image
Front panel discoloration due to metallic interconnections overheating Degradation of power loss in steps over time Failure may cause fire, electric shock, and physical danger image
Multicrystalline Si module delamination Degradation of power loss in steps over time Failure may cause physical danger image
Thin film module delamination Degradation of power loss in steps over time Failure may cause physical danger image
Glass breakage in thin film modules Degradation of power loss in steps over time Failure may cause physical danger image
EVA browning Degradation of power loss linearly over time No effect on safety for slightly browned condition, but as the browning grows faster fire failure may be caused image
Snail trails Degradation of power loss linearly over time Fire failure may be caused image
Back sheet delamination Degradation of power loss linearly over time Fire failure may be caused image


During the whole life cycle of photovoltaic cells, or more precisely modules, it is needed to evaluate whether or not the PV devices have the declared parameters. During the time of use, ie the time of operation, it is necessary to perform not only diagnostics of visible defects but also preventive diagnostics, which helps to detect possible defects in time and thus eliminate economical losses, eg possibly dangerous microcracks. There are many methods for evaluation of parameters in operation phase of a PV device, some of which have support in international standards.

3.1 Diagnosis using radiation

3.1.1 Thermography

When assessing parameters gained through electric measuring, it is necessary to disconnect the cells and especially the modules from the rest of the device and to measure them separately, usually in a laboratory. In case of PV modules, this method can be quite expensive as well as rather time consuming with less accessible installations. If only for this reason, the use of any method, enabling defect detection without the need of disassembly is necessary. Thermography is one such method which is frequently used in diagnostics of faults of PV installations.59 The method works at a principle of detection of thermal irradiation using an appropriate detector. The cells in the module are connected serially. Therefore, when an ideal state is considered where all the cells are identical, the same current flows through them while having the same voltage, ie they have the same short circuit current. When the short circuit current flows through all the cells, the voltage is zero. In case of one faulty cell, the current and voltage proportions in the circuit change as given in.60 Given that the overall voltage in short circuit state has to stay zero, the damaged cell is reverse polarized toward other cells and the voltage in it is the total sum of voltage of other cells in forward direction. In case of the real operation, the cells function on a level of the highest output, but the mechanism of heating of the damaged spots is identical.61 This effect can be caused either by the defect in the cell or by shading. For a reduction in this influence so called bypass diodes are used. These diodes open in case the voltage drop in a reverse direction exceeding their threshold voltage VB. The bypass diode is not used for every single cell in the actual modules but it has an antiparallel connection to a group of serially connected cells (usually 20-24 cells).62 The functionality of the bypass diode is also apparent from the I-V curve, where the “stairs” will occur, which can complicate the detection of MPP, because more than one local extreme is then present at the curve (Figure 3).63

Details are in the caption following the image
Figure 3: Shading on one submodule string

The Shading on one submodule string where shaded irradiance is approximately 150 W/m2 and nonshaded irradiance is approximately 1000 W/m2. The distribution of temperatures can be detected by the use of appropriate equipment. Reading of thermal field used to be done by thermometer (whether they were contact or contactless). These days, thermal cameras are broadly used thanks to the drop in their price. Detection with the thermal camera allows uncovering a wide range of defects as depicted in Table 2.64 However, the disadvantage is that the thermogram does not provide a quantitative assessment of the examined module.65

Table 2. Thermography based fault identification in PV modules64
Observation Detail Reason for failure Electrical characteristics Power loss Safety issue
image Some cells of a module are warmer than other cells due to random distribution of individual cells- (patch work pattern) Incorrectly connected cells Complete module short circuit Constant power loss Fire may occur due to short circuits
image One module in the array is warmer than others Open circuited module Fully functional and normal module Considered as a system failure No effect on safety
image A particular cell in the module is warmer than the other cells Effect of shadowing, defect in cells, delaminated cell Power loss is observed, but it is not necessary for every instant of this failure Power loss increases with mechanical load, thermal cycling, and humidity Extreme conditions may lead to fire
image A particular part of the cell in a module is warmer than the other cells This may be due to broken cell, or interconnection failure Power and form factor reduces drastically Power loss increases with mechanical load, thermal cycling Fire may occur due to failure
image The substring part of the module is remarkably hot Short circuit in cell string because of faulty bypass diode Very high short circuit current and reduction in power <3% of power loss, linearly over time Extreme conditions may lead to fire

3.1.2 Electroluminescence

Another very useful diagnostic tool is electroluminescence (EL). This method is utilized for cells and modules’ defects like cracks (the usual short is than ELCD—Electroluminescence Crack Detection Test), technological defects and other inhomogeneities evaluation. Thus, it can also serve as the visual evaluation of the modules. It works at the principle of electroluminescence radiation detection which is emitted by recombined charge carriers during the radiative recombination process. The equipment should be located in a place with sufficiently low irradiance (the dark room is the best) and the sensor must be a special sensor which allows detection in near IR area (the radiation of PV modules has the wavelength of about 1 μm). The usually used sensors are cooled CCD, CMOS, or InGaAs sensor. The third one has the advantage of a much higher sensitivity in the useful wavelength area, so lower exposition time can be used—in the case of CCD camera, the good EL image takes about 5 minutes, in the case of InGaAs camera, only a few milliseconds.66 The PV module is connected to the current source and the current which should not exceed ISC value flows through it. The intensity of emitted radiation is dependent on this current and on lower current level, different defects will occur. Places, which are affected by some damage, places with higher defects density, respectively, are seen as dark places at the EL images. Such places do not contribute to electricity production. The radiation intensity is then the scale of PV module functionality.67

3.1.3 Photoluminescence

This method is used for PV cells diagnostics. Unlike the EL it does not need the sample with the contact system, because the radiative recombination excitation is stimulated by the strong light impulse, so the method can be used like the control procedure during the manufacturing process, which can be very useful, for example, during layers deposition. The method disadvantage is the need of special sensor (like the EL case) and also much more complicated equipment for radiation excitation.6869

3.1.4 Microplasma luminescence

This method provides the information about shorts inside the structure. Unlike the EL measurement, the module in this case is connected in reverse direction, but the reverse polarization must not exceed the breakdown voltage of the cell.70 The reverse polarization in the areas affected by some defect causes the microplasma occurrence. Microplasma manifests either like the noise or like the light emission. The light emission causes light places in obtained pictures, which means that these pictures are as a matter of fact inverse to the electroluminescent ones.71

3.2 Static characteristics measurement—current-voltage characteristics

Current-Voltage characteristics (I-V curves) measurement method, sometimes also called I-V analysis or Voltammetry, is the most widespread method for PV cells and modules diagnostics. It allows determination of basic PV components parameters. The measurement can be performed either in a lab or directly at the installation, but it is used mostly for the precise measurement under Standard Test Conditions which are specified in international standards like irradiance = 1000 W/m2, and cell temperature 25°C. This can be achieved by using continual solar simulator or by flash solar simulator. Usually, the measurement is performed under flash solar simulator—flash tester, because it is much less complicated and cheaper solar simulator than the continual one and the pulse duration is sufficient for most technologies. During PV component irradiation, the whole I-V curve through the electronic load control is measured. The continual solar simulator is necessary only in the special cases like concentrator modules or solar thermal collectors. For outside measurement, the solar analyzers, which usually allow also connection of external sensors for ambient conditions measurement are used.72 The most problematic part is testing thin films due to season annealing effect and high capacitance.

3.2.1 Effect of season annealing

During measurement of thin-film modules, it was relatively early observed, that the performance and efficiency is very dependent on the history of operation. It means that during the operation, the module has for example MPP of 100 W (under STC). After putting it into the dark storage and measuring it again, only 90 W is measured. It was found, that the temperature and irradiance cause metastable states of PV modules which must be either eliminated or calculated when measured. Season annealing effect covers two phenomena: light-soaking (LS) effect and temperature annealing (TA) effect. The first one is very well-known especially from amorphous silicon-based PV modules, where so called Staebler-Wronski effect occurs.73 Within this effect, optically excited carriers are breaking weak Si-Si bonds leaving them free, so it means that the recombination centers are then created and the carrier’s lifetime is lowered. This drop occurs during the first several hundreds of hours of operation and can reach up to 30% efficiency decrease. For suppression, the hydrogen is used and also tandem cells show lower level of LS effect. The other thin-film technologies also display similar behavior. In many experiments it has been detected, that CdTe PV modules often have a significant increase in device performance in the range from a few percent up to 10 % within the first hours of light soaking.74 This performance decrease can be also achieved by applying a forward biased current in the dark conditions. The CdTe modules biggest producer has its own methodology for measuring their modules, which is based on the presumption that when the module is measured between April and September, its performance should be stable if measured before the third day of storage. Positive influence of light soaking can be observed also in the CIGS modules. Unlike the CdTe modules, there is still no satisfactory theory explaining this effect in CIGS modules. Temperature annealing always has a good influence on the thin-film modules performance. When higher temperature is applied, even Staebler-Wronski effect can be recovered.

3.2.2 Capacitance effects

The modules with higher capacitance, eg thin-film modules, can be easily wrongly evaluated when measured by the flash tester. This is caused by charging or discharging the capacity. When the module is measured from the short circuit state, the measured performance value can be lower than the real one—the capacity is charged. During the reverse measurement—from open circuit state to short circuit state, this capacity is discharged and it can cause the virtual increase in the measured performance. If there are no visible differences between forward and reverse regime, the pulse duration can be considered as sufficiently long.75 The pulse duration influence can be eliminated either by a sufficiently long pulse duration or using multiflash measurement or also by the so called “dragon back pulse” tester of company PASAN who developed the special measurement method of high-capacitance modules using one single controlled 10 ms pulse.10 It is observed that most literature on the topic of I-V curves measurements stated that capacitance is the major concern for distortions in output characteristics of solar cell as in.7679 The problem with this behavior is that if it is modeled like the common capacitor, it does nt work. With larger amount of PV cells, the resulting capacitance should be lower, but the opposite situation happens. These effects can be relatively easily simulated using SPICE (parameters of the diode can be edited to achieve PV cell behavior). Unlike the classical diode model, SPICE diode model covers these effects by “borrowing” the Transition Time (TT) parameter from the transistor’s theory. It is very useful, because this explains a lot in the connection with the “strange” behavior of the capacitance inside the structure. During irradiation, the capacitance is in ideal case infinite—the electrodes of the virtual capacitor, it means grounds of P and N type area are infinitely close to each other. Whole area is then full of nonequilibrium carriers. These carriers need some time to leave and create the depletion region again to recover the previous equilibrium state, if the module is put into the dark again. This time, it can be characterized just by the transition time parameter, so the resulting value of serial connection of capacitors inside the module can be imagined more as the connection of batteries and some kind of transition charge. The similar effect can be achieved by changing the capacitance values, but the problem is, that this capacitance is then necessary to change for every pulse duration change, when proper values should be obtained. This is caused just by the fact that the capacitance in this case cannot be represented by the simple capacitor as mentioned above. The different faults that can be identified using deviation of I-V characteristics both at cell level and Module level can be observed in Tables 3 and 4.

Table 3. Cell failure detection using I-V characteristics
  Failure Broken cell interconnect ribbons Cracked cells Short- circuited cells
  Power loss Degradation of power loss in steps over time Degradation of power loss in steps over time Degradation of power loss in steps over time
Characteristic Safety issue Failure may cause fire, electric shock, and physical danger No effect on safety No effect on safety
P max  
I SC image    
V OC image    
R OC image  
R SC image      
  • F, failure condition; NF, no failure.
Table 4. Module failure detection using I-V characteristics
  Failure Bypass diode (short-circuit) Delamination Induced degradation Solder corrosion
Homogenous Heterogenous Potential Light
  Power loss Degradation of Power Loss in steps over Time Degradation of Power Loss which saturates over time Degradation of Power Loss which saturates over time Degradation of power loss linearly over time Degradation of Power Loss in steps over time Degradation of power loss linearly over time
Characteristic Safety issue Failure may cause fire, Electric Shock, and Physical Danger Failure may cause fire, Electric Shock, and Physical Danger Failure may cause fire, Electric Shock, and Physical Danger No effect on safety No effect on safety No effect on safety
P max  
I SC image      
V OC image      
R OC image          
R SC image          
  • F, fault condition; NF, no fault.

3.2.3 Measurement of dark I-V curve

A very interesting part of static characteristics measurement is the dark current measurement. This method is usually used for determination of dominant disorder and its concentration in PV cells, especially silicon ones.

Determination of recombination centers types

Radiative recombination, recombination through the local recombination centers, Auger recombination are possible recombination processes inside the PV cell. In the silicon cells which have indirect band structure, the recombination caused by local recombination centers predominate known as Shockley-Read-Hall theory.80 Local centers are created by dopant atoms whose energy lies inside the cell band gap. Such dopant atoms function as local recombination centers, also called traps, which capture generated electrons during their transit back to valence band. During their passage, these electrons cannot be thermally excited back to conduction band—at first, they recombine at the local center, then they go back to the valence band, so consequently, the resulting recombination significantly increases. Therefore, it’s important to know their character, concentration, and distribution. One of the methods that make to obtain dominant disturbance energy level and its concentration possible is dark current measurement.

Dielectric spectroscopy

The AC parameters can be measured using frequency domain technique—Dielectric spectroscopy. The Dielectric spectroscopy is characterized by the measurement and analysis of some or all impedance related functions of an electronic device. In Dielectric spectroscopy, the complex impedance Z(ω) = R(ω) + jX(ω) of a device is measured directly within a large range of frequencies. A purely sinusoidal voltage with varying frequency is applied to the terminals of the device under test and the phase shift and amplitude of the voltage and current signals are measured. The ratio between the applied voltage and the resultant current is calculated and this gives the impedance Z(ω) of the device under test. The plotting of R(ω) and X(ω) on a complex plane, in function of the varying frequency, gives the impedance spectrum of the device.8283 In the photovoltaic area, this method has been in use for many years, mainly for AC parameters estimation such as RP determination,7781 but in78 the usage for the donor concentration determination is presented as well. For measurement, the LCR meter is usually used.

Cole-Cole diagram

It is the graphical representation of X(ω) vs R(ω) of the solar cell, and it gives the impedance arcs from which the different AC parameters can be extracted. The construction of Cole-Cole diagram allows simple evaluation of parameters and also the quality of the tested cell, module respectively.84 The problem of measurement of the Cole-Cole diagrams lies in many factors that can influence the resulting curves. The Cole-Cole circle diameters correspond with the PV cell parameters, but also with other aspects like irradiance level and its spectrum, DC biasing, test signal level and cable length. The most problematic parameter is the capacitance which is the representation of physical capacitances inside the structure, particularly space charge capacitance of depletion region (transition capacitance CT) and capacitance due to minority carrier’s oscillation in response to the AC signal (diffusion capacitance CD). Both capacitances represent relatively difficult problem because it shows strong dependence on the voltage and injection level.


The basic diagnostics of large PV systems is based on data analysis. Usual configuration of large systems consists of string inverters with the middle power (about 5-10 kW) because of the problems with local shading caused by clouds movements. The problem of data evaluation is out of range of this work and is described in many publications.8592 When the suspicion on the wrong component is detected, the concrete strings are then evaluated using other methods. In general, thermocouples are used to measure the temperature, and for measuring the temperature of PV module at pixel level thousands of thermocouples are required. While measuring the thermal status at a minute level the thermocouples may object the thermal characteristics or properties. Utilizing thermal cameras, heat signatures underneath any system can be identified. When people hear the term thermal imaging or infrared, they think about finding bad or high temperatures in modules. But in reality, the thermal imaging can do much more. Thermographic inspections have been used for over 30 years as components in both preventive and predictive maintenance programs. The advantage with the thermal image is that it can provide a wealthy information that can improve the efficiency, productivity, and safety. The thermal image of a PV module with hotspot phenomenon is depicted in Figure 4.93 The infrared image depicts the recorded thermal patterns and temperature values across a module with hotspot. This conceptualizes that all objects above absolute zero or zero kelvin emit infrared radiation. When observed with a human eye these infrared energy signals are unnoticed, it is detectable only with the help of a thermal imager. The thermal imager converts the invisible infrared imaging into monochrome or multicolored image, which represents the apparent thermal patterns across the surface of the panel. By looking at the image it can realized that the lighter colors, the whites, the yellows represent a warmer temperature and the blues and blacks represents cooler temperature; these help us in identifying the quality analysis of the module. Using a calibrated camera, temperature values or radiation values from the image can be extracted with the help of mesh analysis. The developed mesh analysis breaks the thermal image into pixels or detectors, from where temperature information can be extracted.

Details are in the caption following the image
Figure 4: Thermography from the air hotspot on a photovoltaic module

4.1 Working with thermal images

Primarily there are two ways to do the thermal analysis: Right on the camera and Information based. Some sought of thermal imaging cameras that almost look like camcorders and behave in the similar ways except for the thing that these thermal cameras have thermal sensors and visual sensors. Whenever, a module is tested for thermal image, functions like spot tools or area tools can be applied and temperature information can be extracted right on the camera. The disadvantage of right on the camera thermal analysis is that it is only limited only with spot tools, area tools. In order to overcome these disadvantages information based thermal analysis can be adapted. This procedure is capable of performing all kinds of analysis like spot, area, polygon, free form, and segmentation analysis. The major advantage of information based thermal analysis is that it can work on matrix level and cover up to a large are of PV systems. This can also be implemented in real time monitoring of PV plant and predictive assessment of aging phenomenon. To check the possibility of quantification of thermograms there were many measurements performed at the real system and in a laboratory. PV modules with defects were detected and were subsequently tested by a flash tester (a device for accurate measuring of I-V curves of modules—establishing nameplate parameters of PV modules). To meet standard condition, the flash test time must be <50 ms.94 All acquired results can be found in.95 The obtained I-V characteristics can be utilized for training of intelligent systems which classify the fault based on the type, nature and effect of fault on the power output of the system.

4.2 I-V characteristics-based fault detections

In this section Wavelet-Based Fault Detection technique9699 is proposed which aims at finding the optimum combination of mother wavelets and the number of wavelet decomposition levels that help extracting the most important attributes from the signal, which are needed for fault diagnosis in a PV module. Figure 5 shows the hierarchical structure of the selection of wavelet transform parameters in which three main parameters need to be determined.100 The first parameter is the wavelet functions, the second parameter is the number of decomposition levels, including the approximations and the details as shown in Table 5, and the third parameter is the signal type, including both voltage and/or current signals and their features to be extracted. In general, the fault location in a PV module will be attempted to diagnose from its output voltage waveforms because the output voltages are normally independent from the load and correspond with fault types and locations.

Figure 5: Hierarchical structure of the selection of wavelet transform parameters
Table 5. Properties of wavelet family
Wavelet family General form Members Filters length
Daubechies dbN db1 – db45 2N
Symlets symN Sym2 – sym31 2N
Coiflets coifN coif1 – coif5 6N
Biorthogonal biorNr. Nd bior1.1 – bior6.8 Max (2Nr, 2Nd) +2
Discrete Meyer Dmey 1 102

where N = wavelet order, r = reconstruction, and d = decomposition. Once the required features of the resultant signals were achieved, trained data are obtained by classifying them. In MATLAB there are various classifier system which provide efficient trained data. In our system Multilayer Perceptron Neural Network (MLPNN) is adapted to classify the data obtained for various faults of PV modules. MLPNN is a hybrid system which is used in training the given set of data for multiple outputs to display the type of fault.101 MLPNN deals with the extracted features of the data obtained from the wavelet transform decomposed and filtered signal.102103 In MLPNN, the data are adapted in the tabular from the workspace which is obtained from the wavelet transforms, thermography, and flash tests as shown in Figure 6.

Details are in the caption following the image
Figure 6: Flow diagram for wavelet neural network fault detection and classification

By applying DWT to the output voltages of the PV modules under normal and fault operating conditions, the energy and power of the signals is obtained. The DWT contains an extensive library of the wavelet basic functions, which makes this transform suitable for transient analysis and, hence, provides time-frequency spectrum at different resolutions. The advantage of computational time for DWT as shown in Table 6, makes it more adaptable. It is observed that for various faults, the features of the system vary and the energy of signal is more in comparison with normal condition. These constraints for output of faulted modules help us in detecting the type of fault and localizing it in a stipulated time.

Table 6. Computational time of transform analysis104
Technique Stationary wavelet transform Discrete wavelet transform Hilbert-Huang transform Continuous wavelet transform Wigner-Ville distribution
Computational time (s) 0.1954 0.0049 0.2410 0.2415 0.0807

The developed diagnosis and localizing method are implemented in three steps namely, Identifying the characteristics of fault, extraction of features, and classification action. In classification process the neural networks are trained with the operating data of PV modules under various faults and working conditions; there by indicating them with predefined binary codes. These binary codes can be used with the intelligent fault diagnosis to observe the fault type and its location. The output I-V characteristics of PV modules bearing different faults are illustrated in Figures 711. It can be observed that the signals cannot be directly adapted for fault classification hypothesis, due to correlation with each other. In order to differentiate these signals, a signal transformation technique is adapted. By selecting an appropriate feature extractor, the neural network is provided with adequate yet significant details in the pattern set so that the highest degree of accuracy and performance is achieved. Momoh et al105 explained various signal transformation techniques suitable for training neural network for fault diagnosis. Prior to application of wavelet transform, the VI signal was filtered by applying median filter (a nonlinear digital filter) in order to remove any noise present inside the signal. A Daubechies 3 level (db3) wavelet decomposition was applied for sampling the input signal and captures both the location and frequency information. The output decomposition structure consists of the wavelet decomposition vector c and the bookkeeping vector l, which contains the number of coefficients by level. Figure 12 below depicts the three-level wavelet decomposition of the I-V characteristics of a normal operating condition PV module. Once the signal is filtered and decomposed, in order to reduce the feature dimension, the feature extraction methods are generally implemented at each decomposition level. Various features like energy, Shannon-entropy, power, Signal to noise ratio, Total harmonic distortion, peaks analysis, frequency analysis, and spectrum analysis were used as the feature’s extractors.

Details are in the caption following the image
Figure 7: Delamination of PV module (left), Thermal image of delaminated PV module (Middle) and VI characteristics of the observed fault (right)
Details are in the caption following the image
Figure 8: Snail trails on PV module (left), Thermal image of faulted PV module (Middle), and VI characteristics of the observed fault (right)
Details are in the caption following the image
Figure 9: Defective bypass diode in PV module (left), Thermal image of fault effected PV module (Middle) and VI characteristics of the observed fault (right)
Details are in the caption following the image
Figure 10: Broken interconnects of PV module (left), thermal image of broken interconnects PV module (Middle), and VI characteristics of the observed fault (right)
Details are in the caption following the image
Figure 11: Potential induced degradation (left), thermal image of fault (Middle), and VI characteristics of the observed fault (right)
Details are in the caption following the image
Figure 12: Coefficients of three level wavelet decomposition

The mathematical assessment for the feature’s extractors were given described as follows:

Energy of the signal E is defined as:


Where x(n) is the signal whose energy feature is to be extracted.

The Power feature of signal is defined as:


Usually the limits are taken over an infinite time interval.

In addition to power and energy of signal there were various other features that vary from signal to signal. It was obvious that every signal is prone to some sought of noise which may make the signal undetectable. This Noise varies from signal to signal and hence providing a wide scope for differentiating the signals. The ratio of signal to noise power is defined as Signal to Noise Ratio and is calculated from:


*SNR in decibels (dB)


In order to calculate the SNR, root mean square (RMS) of the noise is to be measured. Considering a digitized signal, the RMS can be calculated by squaring each value of the signal, deriving the arithmetic mean of the squared values, and applying square root to the result. Generally, the RMS of a signal represents the average “power” of a signal. To Identify spectral features in signals using wavelet techniques, initially the signal is preprocessed to remove artifacts and Perform wavelet-based time-frequency analysis to identify features. The instantaneous frequencies associated with time series are represented with numerical solutions of the obtained signals, producing the frequency evolution. This allows us to differentiate between normal and faulted modules. Since for faulted modules the frequencies are constant and coincide with the Fourier frequency, resonance channels are identified due to the good accuracy in the assignments of frequencies. Furthermore, the time evolution of the frequencies allows us to detect temporary resonance trapping of faulted modules and their implications in system performance. The Time Frequency analysis, Spectrum analysis, and Spectrogram of I-V characteristics of PV module operating under normal conditions were depicted in Figure 13.

Details are in the caption following the image
Figure 13: Time Frequency analysis, Spectrum analysis, and Spectrogram of VI characteristics of PV module operating under normal conditions

Extracting all the mentioned features for different signals obtained for normal and faulted modules, MLPNN technique is used to classify them, to obtain the trained data which can be used for fault identification as shown in Figure 14.

Details are in the caption following the image
Figure 14: Block representation for developed experimental procedure

The observations reveal that the output voltages are related to the fault locations and fault types. Once all the fault signals were classified, a suitable intelligent technique such as neural network is applied to classify the fault features. The application of neural networks (multilayer perceptron) especially for fault classification as depicted in Figure 14 is observed to be the most advantageous solution. The other advantages with neural networks are that, for a single neuron failure the performance of the network is partially degraded but can still decide by using the remaining neurons hence provide an extra degree of freedom to solve nonlinear problems. The trained data which is obtained by classifying the extracted features is used to detect the fault for a given set of a data. The prediction of fault for a trained data set is shown in Figure 15.

Details are in the caption following the image
Figure 15: Fault detection using MLPNN classifier trained data

The major advantage with the algorithm is the classification accuracy and time taken to detect the fault. The classification accuracy (Cacc) of the system depends on the incorrectly predicted cases (Ipc) and number of test cases (Tc) given by (5).

The total accuracy for a given classification model Cacct for N trails is given by (6):


Where Nf determines the number of folds trails. The fault detection time for a given algorithm is defined as the time elapsed between occurrence of fault and the classification of the fault. Average Fault Detection times were significantly shorter for the modules in the Alarm condition, with an average Fault Detection time of 9 seconds.

The fault detection time is given by (7):


Where, FDT is Fault Detection Time, AFT = Alarm and Flash test time (The flash test time should be <50 ms),94 SPT = Signal processing time and CDT = Classification and Decision time. Times for I-V characteristic processing (SPT) and classification and decision (CDT) were 3 seconds, and 5 seconds respectively. The alarm time depends on the area and number of PV modules installed in a particular area.9499106107 Average fault detection time and efficiency of different wavelets along with various classifiers were presented in the Table 7.

Table 7. Wavelet and classifier algorithm accuracy for fault detection
Classification technique Wavelet Signal Misclassified records Accuracy (%) Confidence intervals (%) Average fault detection time
MLPNN (developed) db3 Voltage 0 100 99.00-100 9 s
Support vector machine108 db3 Voltage 2 98.55 96.55-99.42 25 s
K-nearest neighbour classifier109110 db3 Current 7 98.71 97.30-99.40 45 s
Decision tree classifier111 db1 Voltage and current 126 76.12 70.30-77.80 15 min


A case study involving three different operating conditions of PV modules is carried out in the Advanced Power Electronics Lab of Jamia Millia Islamia-New Delhi, India. Three different PV modules each of rating 250 W (maximum power) which were subjected to normal operation (NO – 001b), dusted operation (DM – 010b), and hotspot (HM – 100b) were tested under standard test conditions. The Performance of the modules is assessed under 25°C cell temperature with an airmass of AM 1.5 and an irradiance of 1000 w/m2. The standard test conditions correspond to the sunlight spectrum which is incident on a clear sunny day with sun facing 37°—tilted and the angle of sun is about 41.81° above the horizon. The thermal images corresponding to various faults and operating modes of the panels were captured using TiS45 infrared camera as shown in Figure 16 and their corresponding I-V Characteristics were obtained. The observed I-V characteristics were tabulated for extracting the features as per the proposed feature extraction procedure. Once the required features are observed, they are divided into three classes and were subjected to training using multilayer perceptron neural network as shown in Table 8.

Details are in the caption following the image
Figure 16: Layout of case study
Table 8. Features extracted
Sample number Features extracted Output (from MLP) Fault status
Energy Power Peak Total harmonic distortion Entropy Signal to noise ratio
1 99.957 142.1571 186.3215 0.155154 0.0899 271.0186 (001) b NO
514 96.432 136.2468 169.425 0.18435 0.3325 289.0177 (010) b DM
1027 98.356 139.346 154.36 0.2286 0.2488 276.8695 (100) b HM
1539 99.993 147.256 156.25 0.2168 0.2675 274.3597 (100) b HM

Once the training data set is obtained, it is transformed to Altera DE2 115 Cyclone IV series FPGA (Field Programmable Gate Array) board for testing of the developed training algorithm. For the testing purpose, the features of a known hotspot fault without any classification were passed through the trained data set and the final classification outputs were observed on the 16 × 2 LCD display mounted on the FPGA. By adapting the Thermal inspection tool for fault diagnosis of PV modules a variety of data mining and analysis techniques may prove useful for understanding module degradation and failure. Conventional diagnostic methods and their limitations were considered and a correlation between conventional and less commonly used diagnostic methods were observed resulting in definition of a new approach in diagnosing and localization of faults. Table 9 depicts the summarized parameters of training, validation, and testing samples and the structure of the MLPNN technique. In short, the results depicted 100% training accuracy and 99% testing and validation accuracy in a very short time of <9 seconds for the developed algorithm.

Table 9. Summarized parameters for the MLPN network
Parameters Wavelet-MLPNN classification process
Training process Testing & validating process
No. of samples used 1539 462
Network structures 5-12-3 5-12-3
Epochs 26 26
No. of unclassified samples 0 2
Performance criteria Mean squared error Mean squared error
Training algorithm Scaled conjugate gradient backpropagation Scaled conjugate gradient backpropagation
Classification accuracy 100% 99%


The work presented in this paper aims to study different types of faults in PV modules and their detection techniques. A state-of-art literature review was introduced in which types of faults in various PV modules were observed and the advantages and disadvantages were highlighted. The drawbacks with various condition monitoring and fault diagnosis techniques with respect to various output characteristics and performance parameters of PV modules were observed. Therefore, the work developed focused on the diagnostic techniques, which are capable of extracting distinguished features that could differentiate between various fault types. The importance of thermal images and their utilization with intelligent techniques is developed for better observation and understanding of the faults. Wavelet transform based signal processing techniques were implemented for feature detection and extraction in the frequency and time domain. The extracted features were trained with neural network systems for classification. The trained data from developed algorithm provided 100% accuracy when adapted to classify a given fault condition. The Average Fault Detection time under Alarm condition is measured to be <9 seconds which is better compared to literature. The effectiveness of the proposed fault classification algorithm is validated in both a simulation platform and a small-scale PV experimental system under real working conditions. From the review and case study, it is learnt that the developed fault classification algorithm is very efficient in terms of training, validation, testing, and fault detection time. Faults that may be hidden in PV array without being noticed can also be detected using the developed algorithm. This makes the system more reliable. The case study verifies the simulation results and proves that the developed algorithm has a better chance to clear the fault. To continue this research, future work may explore new active fault protection solutions, such as how to clear the fault actively, responsively, and safely. Based on the tripping signal generated from the proposed methods, the active fault protection solution should increase the system efficiency, reliability, safety, and fault immunity. Therefore, an integration of active fault protection approaches with the proposed methods would be a nice future research topic.

Leave a Reply

Your email address will not be published. Required fields are marked *

 37 Bùi Hữu Nghĩa, Phường An Hải Bắc, Quận Sơn trà, Thành phố Đà Nẵng

 +84 846 39 19 68